diff --git a/configs/figure5/flat/mlp/seed_9/final.yaml b/configs/figure5/flat/mlp/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..615346796708a31b5a6f638aec86359b6b1f42b3 --- /dev/null +++ b/configs/figure5/flat/mlp/seed_9/final.yaml @@ -0,0 +1,62 @@ +policy: + name: mlp + actor_kwargs: + hidden: + - 1378 + - 1378 + activation: relu + value_hidden: + - 256 + - 256 + value_activation: relu + log_std_init: -1.0 +env: + env_id: Ant-v5 + friction_values: &id001 + - 1.0 + steps_per_context: 1000000000 + cycle: false + append_zero_goal_features: true + stabilization: &id002 + enabled: true + forward_cap_mps: 2.2 + inversion_threshold_degrees: 90.0 + inversion_persistence_steps: 10 + action_scale: 0.9 +resolved_env: + env_id: Ant-v5 + friction_values: *id001 + steps_per_context: 1000000000 + cycle: false + append_zero_goal_features: true + stabilization: *id002 + schedule_seed: 9 +ppo: + total_timesteps: 200000 + n_envs: 4 + n_steps: 512 + seq_len: 64 + n_epochs: 10 + n_minibatches: 4 + gamma: 0.99 + lam: 0.95 + clip_range: 0.2 + vf_clip_range: 0.2 + lr: 0.0001 + vf_coef: 0.5 + ent_coef: 0.0 + max_grad_norm: 0.5 + normalize_reward: true + normalize_observation: true + observation_clip: 10.0 + rollout_heartbeat: 128 + best_checkpoint_mode: stability_velocity + best_checkpoint_window: 100 + best_checkpoint_min_episodes: 20 + best_checkpoint_min_forward_velocity: 1.5 + best_checkpoint_max_high_z_termination_rate: 0.15 + best_checkpoint_max_inverted_completion_rate: 0.05 + best_checkpoint_max_functional_failure_rate: 0.2 +evaluation: + run_robustness: true +seed: 9 diff --git a/configs/figure5/large/adult/seed_0/final.yaml b/configs/figure5/large/adult/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..54f47b6c4875aa056498a6f77696dd7812f0d1bc --- /dev/null +++ b/configs/figure5/large/adult/seed_0/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_1/final.yaml b/configs/figure5/large/adult/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..718c1a99abc06ef96d54df7923becdf6401e9740 --- /dev/null +++ b/configs/figure5/large/adult/seed_1/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_10/final.yaml b/configs/figure5/large/adult/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..14ce3a89828370e8c4635307b3bf190748829a2e --- /dev/null +++ b/configs/figure5/large/adult/seed_10/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_11/final.yaml b/configs/figure5/large/adult/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9f96117045dc316e1f57eac06ec544406bb8dbfb --- /dev/null +++ b/configs/figure5/large/adult/seed_11/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_12/final.yaml b/configs/figure5/large/adult/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f0f43e7b943cc7c12367db790a34a48779365d60 --- /dev/null +++ b/configs/figure5/large/adult/seed_12/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_13/final.yaml b/configs/figure5/large/adult/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2f419c4c01d91d51af699926219b951f941eeb60 --- /dev/null +++ b/configs/figure5/large/adult/seed_13/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_14/final.yaml b/configs/figure5/large/adult/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..63415a3e71db2be38df2af66035c42a2453713ba --- /dev/null +++ b/configs/figure5/large/adult/seed_14/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_15/final.yaml b/configs/figure5/large/adult/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ed69073c392fbc91373adadba5ba6d4e9eb12b88 --- /dev/null +++ b/configs/figure5/large/adult/seed_15/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_16/final.yaml b/configs/figure5/large/adult/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..583d89bee96fe0216c2bc5edfdc40ff68385b883 --- /dev/null +++ b/configs/figure5/large/adult/seed_16/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_17/final.yaml b/configs/figure5/large/adult/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..45738acf5ad1fdba10527dce3f5ec10d072abf14 --- /dev/null +++ b/configs/figure5/large/adult/seed_17/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_18/final.yaml b/configs/figure5/large/adult/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bd07c3d9bb8579682d0c1129700f8c62fe812338 --- /dev/null +++ b/configs/figure5/large/adult/seed_18/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_19/final.yaml b/configs/figure5/large/adult/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7a68ae52963e2612ff7aa681cc6676bc55a3ab51 --- /dev/null +++ b/configs/figure5/large/adult/seed_19/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_2/final.yaml b/configs/figure5/large/adult/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dbed0689f862cc98b41cf1049b5eb94c79fdb01e --- /dev/null +++ b/configs/figure5/large/adult/seed_2/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_3/final.yaml b/configs/figure5/large/adult/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e8e4605e027b1e031bcfeb447132bb4bb633e3a4 --- /dev/null +++ b/configs/figure5/large/adult/seed_3/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_4/final.yaml b/configs/figure5/large/adult/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d0c3e8d487fab1effac3e4678bd8c36ef06bbc2c --- /dev/null +++ b/configs/figure5/large/adult/seed_4/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_5/final.yaml b/configs/figure5/large/adult/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2c6f674b8de24a573e4b7bbec6bf8944856a9d1f --- /dev/null +++ b/configs/figure5/large/adult/seed_5/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_6/final.yaml b/configs/figure5/large/adult/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..01cb37ab86ced455850e673084fb8a005adf5650 --- /dev/null +++ b/configs/figure5/large/adult/seed_6/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_7/final.yaml b/configs/figure5/large/adult/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..083222ce10d560d37599666c8b46d0ae6f977709 --- /dev/null +++ b/configs/figure5/large/adult/seed_7/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_8/final.yaml b/configs/figure5/large/adult/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3ee42f0908af6df7e78141fe7bae50de95d4b3de --- /dev/null +++ b/configs/figure5/large/adult/seed_8/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/adult/seed_9/final.yaml b/configs/figure5/large/adult/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f5a2f0b63403d0c9d3252b05b8bf9c9394a3c9cd --- /dev/null +++ b/configs/figure5/large/adult/seed_9/final.yaml @@ -0,0 +1,176 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_0/final.yaml b/configs/figure5/large/gru/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bd46a85de58f13250b289c96b93dc7b4b01e6ee5 --- /dev/null +++ b/configs/figure5/large/gru/seed_0/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_1/final.yaml b/configs/figure5/large/gru/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c070f9b1828d04fd940efe0c5741f8b577eb3184 --- /dev/null +++ b/configs/figure5/large/gru/seed_1/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_10/final.yaml b/configs/figure5/large/gru/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4e90442a1dbe3d38ecb059ceba75191aab7cb83e --- /dev/null +++ b/configs/figure5/large/gru/seed_10/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_11/final.yaml b/configs/figure5/large/gru/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bf465502c9134bb5a7cedf4ed83886257549345a --- /dev/null +++ b/configs/figure5/large/gru/seed_11/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_12/final.yaml b/configs/figure5/large/gru/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a71500efe2ffa96635e69e560df61a9181a54d7f --- /dev/null +++ b/configs/figure5/large/gru/seed_12/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_13/final.yaml b/configs/figure5/large/gru/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..930ff277a6c556f956b0490f40ae8e8baded15a2 --- /dev/null +++ b/configs/figure5/large/gru/seed_13/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_14/final.yaml b/configs/figure5/large/gru/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cb3cd47428d4627687d27626af90b88a25f9617f --- /dev/null +++ b/configs/figure5/large/gru/seed_14/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_15/final.yaml b/configs/figure5/large/gru/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cc6ef1c5b79ccd501dfc235e8ead32e8dc571729 --- /dev/null +++ b/configs/figure5/large/gru/seed_15/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_16/final.yaml b/configs/figure5/large/gru/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2a5c471ddbfb2d90b59ecef3433d36d26f0025bf --- /dev/null +++ b/configs/figure5/large/gru/seed_16/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_17/final.yaml b/configs/figure5/large/gru/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9708b8884235affbe401ba69bac2638183188517 --- /dev/null +++ b/configs/figure5/large/gru/seed_17/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_18/final.yaml b/configs/figure5/large/gru/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b2f54f6c7763989333bd045ed5862e459a3b6f99 --- /dev/null +++ b/configs/figure5/large/gru/seed_18/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_19/final.yaml b/configs/figure5/large/gru/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..993b984e3804cb994d3620e42b475684c0b5e912 --- /dev/null +++ b/configs/figure5/large/gru/seed_19/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_2/final.yaml b/configs/figure5/large/gru/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..45a2b24c257c8dd255db7c37101ffe275ef85773 --- /dev/null +++ b/configs/figure5/large/gru/seed_2/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_3/final.yaml b/configs/figure5/large/gru/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..48e69183fe139e2aeeda96741527eaf4d53ced70 --- /dev/null +++ b/configs/figure5/large/gru/seed_3/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_4/final.yaml b/configs/figure5/large/gru/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..91e7ada35829f7d56cc49a19412f7401db80c5a9 --- /dev/null +++ b/configs/figure5/large/gru/seed_4/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_5/final.yaml b/configs/figure5/large/gru/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0462e33bba9dea0168d9c03e31a6eaf91155295f --- /dev/null +++ b/configs/figure5/large/gru/seed_5/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_6/final.yaml b/configs/figure5/large/gru/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..495594d8268e7d8a406b72a7ea8330cb6cb9b30e --- /dev/null +++ b/configs/figure5/large/gru/seed_6/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_7/final.yaml b/configs/figure5/large/gru/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3f30f286e9b8c3c63c0a141fd6bb5cb53a6254c3 --- /dev/null +++ b/configs/figure5/large/gru/seed_7/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_8/final.yaml b/configs/figure5/large/gru/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..13f3786b2a09c6acd3bd6ca6ddda3cfc9eaa1cd6 --- /dev/null +++ b/configs/figure5/large/gru/seed_8/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/gru/seed_9/final.yaml b/configs/figure5/large/gru/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a4f4702b593d4e77e0b26ef96c005668c50d58fe --- /dev/null +++ b/configs/figure5/large/gru/seed_9/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_0/final.yaml b/configs/figure5/large/lstm/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8267510eac88b297e03151134ba6f38e70ff0e2f --- /dev/null +++ b/configs/figure5/large/lstm/seed_0/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_1/final.yaml b/configs/figure5/large/lstm/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..553e6b9719c679679c7c558a97bbf15ed8caff2e --- /dev/null +++ b/configs/figure5/large/lstm/seed_1/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_10/final.yaml b/configs/figure5/large/lstm/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e95fa63836f251316ec2ade8339da3a27a46be09 --- /dev/null +++ b/configs/figure5/large/lstm/seed_10/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_11/final.yaml b/configs/figure5/large/lstm/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..51607bfd9734fbf07ab0798b1bc07fb26fd75c1b --- /dev/null +++ b/configs/figure5/large/lstm/seed_11/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_12/final.yaml b/configs/figure5/large/lstm/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..03790184e38013151b2d2fc2d85c01deec369f9b --- /dev/null +++ b/configs/figure5/large/lstm/seed_12/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_13/final.yaml b/configs/figure5/large/lstm/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..97b1dc3d4e81fe5753156f728a410af3a169dcde --- /dev/null +++ b/configs/figure5/large/lstm/seed_13/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_14/final.yaml b/configs/figure5/large/lstm/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9ef610deaba4afac6db836ee931f6aa01c18192f --- /dev/null +++ b/configs/figure5/large/lstm/seed_14/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_15/final.yaml b/configs/figure5/large/lstm/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4f73bdf70afa41991ee721a747da462596150a91 --- /dev/null +++ b/configs/figure5/large/lstm/seed_15/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_16/final.yaml b/configs/figure5/large/lstm/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cf426383956f1963b5e1f85a2326c3c57672b5da --- /dev/null +++ b/configs/figure5/large/lstm/seed_16/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_17/final.yaml b/configs/figure5/large/lstm/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5b45215ca66ed9519ff6496c3daf46653688ea65 --- /dev/null +++ b/configs/figure5/large/lstm/seed_17/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_18/final.yaml b/configs/figure5/large/lstm/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c427428a623b57041478d2fa62b1f1e347cec2ad --- /dev/null +++ b/configs/figure5/large/lstm/seed_18/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_19/final.yaml b/configs/figure5/large/lstm/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e9d18589833512c9b175ab53c7c63a424edad13e --- /dev/null +++ b/configs/figure5/large/lstm/seed_19/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_2/final.yaml b/configs/figure5/large/lstm/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5318d31556dd3a07d67f54b91c6b39d37a7d2855 --- /dev/null +++ b/configs/figure5/large/lstm/seed_2/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_3/final.yaml b/configs/figure5/large/lstm/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..29109db2784b49f7cfffae592935a766b3db7819 --- /dev/null +++ b/configs/figure5/large/lstm/seed_3/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_4/final.yaml b/configs/figure5/large/lstm/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..372c4489d7ba51b7daaf62e731847eca17f20dfe --- /dev/null +++ b/configs/figure5/large/lstm/seed_4/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_5/final.yaml b/configs/figure5/large/lstm/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1741201ed6d0cb532e8066785f89b1dec47b9498 --- /dev/null +++ b/configs/figure5/large/lstm/seed_5/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_6/final.yaml b/configs/figure5/large/lstm/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1c98b23fd68235a926ed7c0b4461c9835e8024fa --- /dev/null +++ b/configs/figure5/large/lstm/seed_6/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_7/final.yaml b/configs/figure5/large/lstm/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a76d9c2f7ab14990220c191b062e7248df0552e1 --- /dev/null +++ b/configs/figure5/large/lstm/seed_7/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_8/final.yaml b/configs/figure5/large/lstm/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cdf3202e076bf974fadc753dc46fa277a2115b8b --- /dev/null +++ b/configs/figure5/large/lstm/seed_8/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/lstm/seed_9/final.yaml b/configs/figure5/large/lstm/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..59e6b68b895687ef741706a92273078c39cbcc43 --- /dev/null +++ b/configs/figure5/large/lstm/seed_9/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_0/final.yaml b/configs/figure5/large/mlp/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3ced2406eaf47dfbdce72f4491799eba2c1b7d6a --- /dev/null +++ b/configs/figure5/large/mlp/seed_0/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_1/final.yaml b/configs/figure5/large/mlp/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e17de7f16bb505927998c685d2a5adf54a6f4afa --- /dev/null +++ b/configs/figure5/large/mlp/seed_1/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_10/final.yaml b/configs/figure5/large/mlp/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..709881d93a2acf9acf4d687e194cd2ec92b62559 --- /dev/null +++ b/configs/figure5/large/mlp/seed_10/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_11/final.yaml b/configs/figure5/large/mlp/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..12e4b08d2c2150353755b1028d23b61be370597f --- /dev/null +++ b/configs/figure5/large/mlp/seed_11/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_12/final.yaml b/configs/figure5/large/mlp/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2a148946e8506aaf55651612708b895788a58e69 --- /dev/null +++ b/configs/figure5/large/mlp/seed_12/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_13/final.yaml b/configs/figure5/large/mlp/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5619ed7863a81b750e5aa774358f902e2d39e4cc --- /dev/null +++ b/configs/figure5/large/mlp/seed_13/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_14/final.yaml b/configs/figure5/large/mlp/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4e8361722bafcb51d34d04491dd98ee044c06934 --- /dev/null +++ b/configs/figure5/large/mlp/seed_14/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_15/final.yaml b/configs/figure5/large/mlp/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ebbb874d237aa3ba9938697d18c68c39ffd8697b --- /dev/null +++ b/configs/figure5/large/mlp/seed_15/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_16/final.yaml b/configs/figure5/large/mlp/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1089acb6c4de43cb4b46fb2c0699563786f23aee --- /dev/null +++ b/configs/figure5/large/mlp/seed_16/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_17/final.yaml b/configs/figure5/large/mlp/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..81d0eeb2bb4b38bf6c96c21be4bc825dfca66773 --- /dev/null +++ b/configs/figure5/large/mlp/seed_17/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_18/final.yaml b/configs/figure5/large/mlp/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..76a83550ce990167627147e61ee6648d1ff8cd45 --- /dev/null +++ b/configs/figure5/large/mlp/seed_18/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_19/final.yaml b/configs/figure5/large/mlp/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a6bc20bb22f2a755160d20b52353dcfd65a5d8d1 --- /dev/null +++ b/configs/figure5/large/mlp/seed_19/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_2/final.yaml b/configs/figure5/large/mlp/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..901d8b15665d88ab1cf791c0d24037108eb2d775 --- /dev/null +++ b/configs/figure5/large/mlp/seed_2/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_3/final.yaml b/configs/figure5/large/mlp/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..aeae7cf5576313f2d8d62003af84ab893681c6e6 --- /dev/null +++ b/configs/figure5/large/mlp/seed_3/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_4/final.yaml b/configs/figure5/large/mlp/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..40925127288e3283b2a6b76101ddacfb3bd30ca2 --- /dev/null +++ b/configs/figure5/large/mlp/seed_4/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_5/final.yaml b/configs/figure5/large/mlp/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..377110e001fa8d34468ed84859cc3da7190004ee --- /dev/null +++ b/configs/figure5/large/mlp/seed_5/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_6/final.yaml b/configs/figure5/large/mlp/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b44a7bc0298f6b5c7a4a3d71f1cf1563a0a0261c --- /dev/null +++ b/configs/figure5/large/mlp/seed_6/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_7/final.yaml b/configs/figure5/large/mlp/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ff2e1326466c1377b6a11f67bd8b93b4c7e20dbf --- /dev/null +++ b/configs/figure5/large/mlp/seed_7/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_8/final.yaml b/configs/figure5/large/mlp/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fdfcca120e64fb985a4474b105ae8c01dba2d51a --- /dev/null +++ b/configs/figure5/large/mlp/seed_8/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/mlp/seed_9/final.yaml b/configs/figure5/large/mlp/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2ed1b8dc96fc18789b1940159cd9cb66ecdfe495 --- /dev/null +++ b/configs/figure5/large/mlp/seed_9/final.yaml @@ -0,0 +1,145 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_0/final.yaml b/configs/figure5/large/reservoir/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..75c24b135bf6461dccb9ee4228974f2ab763fb30 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_0/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_1/final.yaml b/configs/figure5/large/reservoir/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a81eb5bca1cca922db14f9fcf9d24d5762a3f404 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_1/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_10/final.yaml b/configs/figure5/large/reservoir/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fc5944ca50363fe402c56beaa0d7568353f4c84a --- /dev/null +++ b/configs/figure5/large/reservoir/seed_10/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_11/final.yaml b/configs/figure5/large/reservoir/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..780cbd2209ee52cca3efb23d069e8949eeeda3b9 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_11/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_12/final.yaml b/configs/figure5/large/reservoir/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..98bbc49bcfe0e8d7e230378f8722abdb0f8ea7aa --- /dev/null +++ b/configs/figure5/large/reservoir/seed_12/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_13/final.yaml b/configs/figure5/large/reservoir/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..89ef5bd76e9d0986b6849a74a92185795005e61c --- /dev/null +++ b/configs/figure5/large/reservoir/seed_13/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_14/final.yaml b/configs/figure5/large/reservoir/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ea059ef0e3e126da9afc116eedd7ed590ea761c2 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_14/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_15/final.yaml b/configs/figure5/large/reservoir/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dda2bfec8d620cdd15839bcb05dd69c7fd36a538 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_15/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_16/final.yaml b/configs/figure5/large/reservoir/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..333057bc4ec920287048193f33ca822ed0f187e3 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_16/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_17/final.yaml b/configs/figure5/large/reservoir/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3aae61e45da740dc6c719b474915f8e3480a8ac2 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_17/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_18/final.yaml b/configs/figure5/large/reservoir/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e32de05f0c3afe10a92617f4a8035bb72ae0c2c3 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_18/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_19/final.yaml b/configs/figure5/large/reservoir/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7430c273a5b834fc5acda98ec75fa6efb743e7a2 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_19/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_2/final.yaml b/configs/figure5/large/reservoir/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b97ffb3f3a7fb18116eaee54531c2fe2798552ba --- /dev/null +++ b/configs/figure5/large/reservoir/seed_2/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_3/final.yaml b/configs/figure5/large/reservoir/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c7e5a81b560fcd090d87db6602e246672e959b33 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_3/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_4/final.yaml b/configs/figure5/large/reservoir/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0ff643a7951838c8dc95019990b5d0b620e89d64 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_4/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_5/final.yaml b/configs/figure5/large/reservoir/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..013a5552a2d6ed8f552ffbabfdd8fcd24f7a4b16 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_5/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_6/final.yaml b/configs/figure5/large/reservoir/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..24abb8bd22a30e2f6b2aff339e838b5a61de893c --- /dev/null +++ b/configs/figure5/large/reservoir/seed_6/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_7/final.yaml b/configs/figure5/large/reservoir/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9956d9fad6f295731d605b9f175ce852afe686df --- /dev/null +++ b/configs/figure5/large/reservoir/seed_7/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_8/final.yaml b/configs/figure5/large/reservoir/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..146de6880d1b0dfb6a2ff39262280fdfea163342 --- /dev/null +++ b/configs/figure5/large/reservoir/seed_8/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir/seed_9/final.yaml b/configs/figure5/large/reservoir/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4d6248aecc0764c6e33b5993681c020e3bfa24dd --- /dev/null +++ b/configs/figure5/large/reservoir/seed_9/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_0/final.yaml b/configs/figure5/large/reservoir_ro/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5b92cb21dd176d5e2d916222fe86fa4f5286ca5e --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_0/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_1/final.yaml b/configs/figure5/large/reservoir_ro/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..36eb71f02e30c0c9b6baaedb720800b0337b4b99 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_1/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_10/final.yaml b/configs/figure5/large/reservoir_ro/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5dd976b8016fe6cffb9dd1b8b481c28343d2d6e0 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_10/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_11/final.yaml b/configs/figure5/large/reservoir_ro/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..774c5e68c377765210cc9af6ce7c842b7535d3a3 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_11/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_12/final.yaml b/configs/figure5/large/reservoir_ro/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..82c299abf7a5c024b3795bb9387d4cd3bfa76b9b --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_12/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_13/final.yaml b/configs/figure5/large/reservoir_ro/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3aafbd19875bd402233053ee9af7703602b3d5c0 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_13/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_14/final.yaml b/configs/figure5/large/reservoir_ro/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ce79ee128b934291c5ade521716e9308ed9eecd4 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_14/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_15/final.yaml b/configs/figure5/large/reservoir_ro/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dbf56127cba786144ed6b87b85f24b09b75aeebd --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_15/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_16/final.yaml b/configs/figure5/large/reservoir_ro/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3f2a42c90237fdc04b349a1570e9e1e27920c3ed --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_16/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_17/final.yaml b/configs/figure5/large/reservoir_ro/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2319f067ddafef41d45167ebb1b4764595cbe9c2 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_17/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_18/final.yaml b/configs/figure5/large/reservoir_ro/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e17634c12323ce3e850f8ca41217300271258e05 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_18/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_19/final.yaml b/configs/figure5/large/reservoir_ro/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8f52544178f790af75336ad24b7b2043909f5af5 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_19/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_2/final.yaml b/configs/figure5/large/reservoir_ro/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0901b4a6fe52822bc991c34bb199c93ec8182ba7 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_2/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_3/final.yaml b/configs/figure5/large/reservoir_ro/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e3657b8ccf061aae99adf804752a89af97d33212 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_3/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_4/final.yaml b/configs/figure5/large/reservoir_ro/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7eb12ad19bd9817989db0441fb742dfeff7446cc --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_4/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_5/final.yaml b/configs/figure5/large/reservoir_ro/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0ad66ef7797f3acc4ce7cb50bb6094304cbe30fb --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_5/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_6/final.yaml b/configs/figure5/large/reservoir_ro/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..006dc898f377d265193db3144c1f89dfd1c957db --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_6/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_7/final.yaml b/configs/figure5/large/reservoir_ro/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4279bdb59f68c72e46d7f9510fab2bba8d7c100d --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_7/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_8/final.yaml b/configs/figure5/large/reservoir_ro/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..03276437eef018671001556e90d634fa6741d54c --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_8/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/reservoir_ro/seed_9/final.yaml b/configs/figure5/large/reservoir_ro/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dafb63985c90fcb3a4a10461dfeb43db1819d0f4 --- /dev/null +++ b/configs/figure5/large/reservoir_ro/seed_9/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_0/final.yaml b/configs/figure5/large/transformer/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..48aaa2a9bfbdf4b3e975cba59a96ba5e4e6c6e55 --- /dev/null +++ b/configs/figure5/large/transformer/seed_0/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_1/final.yaml b/configs/figure5/large/transformer/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3f6874a9e0fedabca51db44a562b9a058f4f832d --- /dev/null +++ b/configs/figure5/large/transformer/seed_1/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_10/final.yaml b/configs/figure5/large/transformer/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..14b1c94bc2808c64b5623b419de530dc2fa12861 --- /dev/null +++ b/configs/figure5/large/transformer/seed_10/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_11/final.yaml b/configs/figure5/large/transformer/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8fdc08a34d6c79d815dcdf0079abc62a4c1bbd4e --- /dev/null +++ b/configs/figure5/large/transformer/seed_11/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_12/final.yaml b/configs/figure5/large/transformer/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e2e4a36e25c5f4127af81d7bf78145cb70dbaf7a --- /dev/null +++ b/configs/figure5/large/transformer/seed_12/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_13/final.yaml b/configs/figure5/large/transformer/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9c316664b25d419e5b273cf9abe60be2a11c8893 --- /dev/null +++ b/configs/figure5/large/transformer/seed_13/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_14/final.yaml b/configs/figure5/large/transformer/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..73ffa3f1cdf9e86d0c1bcda5ed23de1072ea0f2d --- /dev/null +++ b/configs/figure5/large/transformer/seed_14/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_15/final.yaml b/configs/figure5/large/transformer/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4c414324375f2927776e91636f0db81fba755df8 --- /dev/null +++ b/configs/figure5/large/transformer/seed_15/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_16/final.yaml b/configs/figure5/large/transformer/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..da1c7c0d07b454038fd46d9c8c8b0e0d9f031714 --- /dev/null +++ b/configs/figure5/large/transformer/seed_16/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_17/final.yaml b/configs/figure5/large/transformer/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d1a052e6fa8a8d5c129cd250c1324e9bc9b045c6 --- /dev/null +++ b/configs/figure5/large/transformer/seed_17/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_18/final.yaml b/configs/figure5/large/transformer/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8ed81491a7385d080f2a8ea4b0fec16dd9d6c424 --- /dev/null +++ b/configs/figure5/large/transformer/seed_18/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_19/final.yaml b/configs/figure5/large/transformer/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7c811639fcf7af8cba5fe8cb08fb791a2254c9e2 --- /dev/null +++ b/configs/figure5/large/transformer/seed_19/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_2/final.yaml b/configs/figure5/large/transformer/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..706a018bca8cbfa7b523d92b15fdf87ed8d667ea --- /dev/null +++ b/configs/figure5/large/transformer/seed_2/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_3/final.yaml b/configs/figure5/large/transformer/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9f598a3ecd16a747781f7a75545b2329b62e8edf --- /dev/null +++ b/configs/figure5/large/transformer/seed_3/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_4/final.yaml b/configs/figure5/large/transformer/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..064a9a8a6cff852ab276c070173ae10bc76ad34e --- /dev/null +++ b/configs/figure5/large/transformer/seed_4/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_5/final.yaml b/configs/figure5/large/transformer/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..91957308e712275a29413f70d45054cdc9d8e425 --- /dev/null +++ b/configs/figure5/large/transformer/seed_5/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_6/final.yaml b/configs/figure5/large/transformer/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f4f08b5772fe51cb92a2147900532f94bac4be91 --- /dev/null +++ b/configs/figure5/large/transformer/seed_6/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_7/final.yaml b/configs/figure5/large/transformer/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..52263963b90c3d48c9ea452cc738fe344721700a --- /dev/null +++ b/configs/figure5/large/transformer/seed_7/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_8/final.yaml b/configs/figure5/large/transformer/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..17d4e5de5ffbbed7a9bf4809bcf22727823f07be --- /dev/null +++ b/configs/figure5/large/transformer/seed_8/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/large/transformer/seed_9/final.yaml b/configs/figure5/large/transformer/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eea964bb994885d117cd5f14d3f0ee4ef13633ec --- /dev/null +++ b/configs/figure5/large/transformer/seed_9/final.yaml @@ -0,0 +1,141 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 60.0 +policy: + name: transformer + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -18.0 + - 12.0 + - - -14.0 + - 12.0 + - - -10.0 + - 12.0 + - - -6.0 + - 12.0 + - - 2.0 + - 12.0 + - - 6.0 + - 12.0 + - - 10.0 + - 12.0 + - - 14.0 + - 12.0 + - - 18.0 + - 12.0 + - - -18.0 + - 8.0 + - - -6.0 + - 8.0 + - - 2.0 + - 8.0 + - - 10.0 + - 8.0 + - - 18.0 + - 8.0 + - - -18.0 + - 4.0 + - - -14.0 + - 4.0 + - - -10.0 + - 4.0 + - - -6.0 + - 4.0 + - - -2.0 + - 4.0 + - - 2.0 + - 4.0 + - - 10.0 + - 4.0 + - - 14.0 + - 4.0 + - - 18.0 + - 4.0 + - - -18.0 + - 0.0 + - - 2.0 + - 0.0 + - - 18.0 + - 0.0 + - - -18.0 + - -4.0 + - - -14.0 + - -4.0 + - - -6.0 + - -4.0 + - - 2.0 + - -4.0 + - - 6.0 + - -4.0 + - - 10.0 + - -4.0 + - - 14.0 + - -4.0 + - - 18.0 + - -4.0 + - - -14.0 + - -8.0 + - - -6.0 + - -8.0 + - - 2.0 + - -8.0 + - - 10.0 + - -8.0 + - - -18.0 + - -12.0 + - - -14.0 + - -12.0 + - - -6.0 + - -12.0 + - - -2.0 + - -12.0 + - - 2.0 + - -12.0 + - - 10.0 + - -12.0 + - - 14.0 + - -12.0 + - - 18.0 + - -12.0 + goal_router_codebook_latch_radius: 0.45 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Large_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 9420000 + evaluation: 9410000 diff --git a/configs/figure5/medium/adult/seed_0/final.yaml b/configs/figure5/medium/adult/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..764c2e660f55aeb3452df1616a6ffc962fef9dc3 --- /dev/null +++ b/configs/figure5/medium/adult/seed_0/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 0 + evaluation: 120000 diff --git a/configs/figure5/medium/adult/seed_1/final.yaml b/configs/figure5/medium/adult/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eb230c7fae1291ec62f2fef0a33ea68814314819 --- /dev/null +++ b/configs/figure5/medium/adult/seed_1/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 0 + evaluation: 120000 diff --git a/configs/figure5/medium/adult/seed_10/final.yaml b/configs/figure5/medium/adult/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5ce1d4bc368d2b2689e75031f60859d90b4dfcb7 --- /dev/null +++ b/configs/figure5/medium/adult/seed_10/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_11/final.yaml b/configs/figure5/medium/adult/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1d020b991ebefb3f207d0c787e89208f48eecfbe --- /dev/null +++ b/configs/figure5/medium/adult/seed_11/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_12/final.yaml b/configs/figure5/medium/adult/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..95a654c3f4a8852b8583474e1eee60ac9c5e6ba7 --- /dev/null +++ b/configs/figure5/medium/adult/seed_12/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_13/final.yaml b/configs/figure5/medium/adult/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6bc6c98e7636910f9d68bc035e09e13096f20922 --- /dev/null +++ b/configs/figure5/medium/adult/seed_13/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_14/final.yaml b/configs/figure5/medium/adult/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9e30501ef02214354b4f06cbc20e835b111d2b07 --- /dev/null +++ b/configs/figure5/medium/adult/seed_14/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_15/final.yaml b/configs/figure5/medium/adult/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..76a2332c50370391b6ba43b119de36b1d6603cec --- /dev/null +++ b/configs/figure5/medium/adult/seed_15/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_16/final.yaml b/configs/figure5/medium/adult/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..728521aabf556e02dbbebcb32c62c3c972c866b7 --- /dev/null +++ b/configs/figure5/medium/adult/seed_16/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_17/final.yaml b/configs/figure5/medium/adult/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3b62b31a9d6a83b2eb551d29e24c2a42b132e516 --- /dev/null +++ b/configs/figure5/medium/adult/seed_17/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_18/final.yaml b/configs/figure5/medium/adult/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c3809a5617f18a2fa5f0e86125a3852735e39e72 --- /dev/null +++ b/configs/figure5/medium/adult/seed_18/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_19/final.yaml b/configs/figure5/medium/adult/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b9d27197637b712c1c07dbaaf83cd4a48186a1be --- /dev/null +++ b/configs/figure5/medium/adult/seed_19/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_2/final.yaml b/configs/figure5/medium/adult/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bdb82272236a9829df5db7c334fddaf2d8668968 --- /dev/null +++ b/configs/figure5/medium/adult/seed_2/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 0 + evaluation: 120000 diff --git a/configs/figure5/medium/adult/seed_3/final.yaml b/configs/figure5/medium/adult/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ff5ec135330eee588b34be46cb108ddd7a1a9fac --- /dev/null +++ b/configs/figure5/medium/adult/seed_3/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 0 + evaluation: 120000 diff --git a/configs/figure5/medium/adult/seed_4/final.yaml b/configs/figure5/medium/adult/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c31cef37077496cbcf745113ef6ba55f102daf2b --- /dev/null +++ b/configs/figure5/medium/adult/seed_4/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 0 + evaluation: 120000 diff --git a/configs/figure5/medium/adult/seed_5/final.yaml b/configs/figure5/medium/adult/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a2859903231cda5b427f8f2f503a4b713d93defd --- /dev/null +++ b/configs/figure5/medium/adult/seed_5/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_6/final.yaml b/configs/figure5/medium/adult/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1fb2759ae70513d514d09ef6900fc036554c0027 --- /dev/null +++ b/configs/figure5/medium/adult/seed_6/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_7/final.yaml b/configs/figure5/medium/adult/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0000e035e614c242d6b2f1fb25b572e95488a3cc --- /dev/null +++ b/configs/figure5/medium/adult/seed_7/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_8/final.yaml b/configs/figure5/medium/adult/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8a431138f3ea3958bfff9fb84702fc5d5fc7891f --- /dev/null +++ b/configs/figure5/medium/adult/seed_8/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/adult/seed_9/final.yaml b/configs/figure5/medium/adult/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f6661081bf3be95d3bf20675773d9e34d134c1a0 --- /dev/null +++ b/configs/figure5/medium/adult/seed_9/final.yaml @@ -0,0 +1,128 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: connectome_rate + actor_kwargs: + dt: 0.01 + tau_init: 0.02 + use_lora: false + input_target: sensory + goal_adapter_hidden: + - 256 + - 128 + goal_adapter_features: 109 + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +connectome: + species: adult + kind: original + adult_dir: larva/data/adult + chunksize: 250000 + min_syn_count: 1 + normalization: row_l1 + global_gain: 1.0 + sensory_selector: + source: classification + field: flow + include: + - afferent + output_selector: + source: classification + field: super_class + include: + - motor + polarity: + unknown_action: zero + nt_signs: + ACH: 1.0 + GLUT: 1.0 + GABA: -1.0 + DA: 1.0 + SER: 1.0 + OCT: 1.0 +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: goal_adapter_only + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/gru/seed_0/final.yaml b/configs/figure5/medium/gru/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..82dcee1e196eb63a4e88607688d34db935f8a519 --- /dev/null +++ b/configs/figure5/medium/gru/seed_0/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 0 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_1/final.yaml b/configs/figure5/medium/gru/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..adb0ec3c33d51228563aa4108b01ea105c474398 --- /dev/null +++ b/configs/figure5/medium/gru/seed_1/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 100000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_10/final.yaml b/configs/figure5/medium/gru/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6e42a5e03bf5713c7cd8738ca1567448d091a8b3 --- /dev/null +++ b/configs/figure5/medium/gru/seed_10/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 1000000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_11/final.yaml b/configs/figure5/medium/gru/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a932fbe8d3f582578c70309fb6906b6c9fce1fcf --- /dev/null +++ b/configs/figure5/medium/gru/seed_11/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 1100000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_12/final.yaml b/configs/figure5/medium/gru/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bfca2026ecea88b8d7382305ad4482b7a2858139 --- /dev/null +++ b/configs/figure5/medium/gru/seed_12/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 1200000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_13/final.yaml b/configs/figure5/medium/gru/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8f9ae45de7134717196c6be02067168527cf8712 --- /dev/null +++ b/configs/figure5/medium/gru/seed_13/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 1300000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_14/final.yaml b/configs/figure5/medium/gru/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d8185d848e5c12be61fca0ad76d7e6d692830b49 --- /dev/null +++ b/configs/figure5/medium/gru/seed_14/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 1400000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_15/final.yaml b/configs/figure5/medium/gru/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..669cf7798e4eb7324aa912a35e687b3968d59d9a --- /dev/null +++ b/configs/figure5/medium/gru/seed_15/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 1500000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_16/final.yaml b/configs/figure5/medium/gru/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ac4558a58de1f4fe518b8e0c3cb8a38605215468 --- /dev/null +++ b/configs/figure5/medium/gru/seed_16/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 1600000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_17/final.yaml b/configs/figure5/medium/gru/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1cde6fe08bc7845a7fff780b5b3aa24b6bd2ad39 --- /dev/null +++ b/configs/figure5/medium/gru/seed_17/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 1700000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_18/final.yaml b/configs/figure5/medium/gru/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..37e61975a9853c21e10f02708f9d24654549a6c5 --- /dev/null +++ b/configs/figure5/medium/gru/seed_18/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 1800000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_19/final.yaml b/configs/figure5/medium/gru/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b638fa549501b50e577f4ae168aecf6959d6705c --- /dev/null +++ b/configs/figure5/medium/gru/seed_19/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 1900000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_2/final.yaml b/configs/figure5/medium/gru/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..29ba40d74ee6ab1036845f9a5686716b4ed36559 --- /dev/null +++ b/configs/figure5/medium/gru/seed_2/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 200000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_3/final.yaml b/configs/figure5/medium/gru/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..19d16d983f190e7b71d246ebd3a7ea9bdc3eb614 --- /dev/null +++ b/configs/figure5/medium/gru/seed_3/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 300000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_4/final.yaml b/configs/figure5/medium/gru/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dd74eb813716e553e75768c69f2b52c81f8f6c2e --- /dev/null +++ b/configs/figure5/medium/gru/seed_4/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 400000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_5/final.yaml b/configs/figure5/medium/gru/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a82de2ca8e3f0941e673fb56031336780e92acac --- /dev/null +++ b/configs/figure5/medium/gru/seed_5/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 500000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_6/final.yaml b/configs/figure5/medium/gru/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dc14395e2dfaf89bf7fd69fab9f94be69a9744bd --- /dev/null +++ b/configs/figure5/medium/gru/seed_6/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 600000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_7/final.yaml b/configs/figure5/medium/gru/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..676435028c5e9676ee997c729a75e029f39307f3 --- /dev/null +++ b/configs/figure5/medium/gru/seed_7/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 700000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_8/final.yaml b/configs/figure5/medium/gru/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a3c0d110ce1a8d700cf31649bed309adcbe4dcfe --- /dev/null +++ b/configs/figure5/medium/gru/seed_8/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 800000 + evaluation: 9300000 diff --git a/configs/figure5/medium/gru/seed_9/final.yaml b/configs/figure5/medium/gru/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..193ffea03acd602a37dca1ed5669ff703cb6d191 --- /dev/null +++ b/configs/figure5/medium/gru/seed_9/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: gru + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 900000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_0/final.yaml b/configs/figure5/medium/lstm/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2e25f75869bb2bb55178b671b97baaf3903c0bdc --- /dev/null +++ b/configs/figure5/medium/lstm/seed_0/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 0 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_1/final.yaml b/configs/figure5/medium/lstm/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f00faa0987bc43712b3c81fc4da2576c7861ff4d --- /dev/null +++ b/configs/figure5/medium/lstm/seed_1/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 100000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_10/final.yaml b/configs/figure5/medium/lstm/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..08c93f6b3990a1ce1d221b537f0f7e28a9544148 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_10/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 1000000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_11/final.yaml b/configs/figure5/medium/lstm/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dcea860ff4e277e0d2720c9f9afd4cb726c86220 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_11/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 1100000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_12/final.yaml b/configs/figure5/medium/lstm/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..baea539c962f88fbb35bf7b5bf13036db9f8f9db --- /dev/null +++ b/configs/figure5/medium/lstm/seed_12/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 1200000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_13/final.yaml b/configs/figure5/medium/lstm/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6da0f9324baf7aabdee357349388a5054b04c7e0 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_13/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 1300000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_14/final.yaml b/configs/figure5/medium/lstm/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fe1d93142143cc95cc206c8f61f0c5a24bed4ec3 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_14/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 1400000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_15/final.yaml b/configs/figure5/medium/lstm/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ad966d43096b776f44b717659e41d3a3c82319a9 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_15/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 1500000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_16/final.yaml b/configs/figure5/medium/lstm/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ab0f9213bf581b0ed9e47da5a42f674fb091b18e --- /dev/null +++ b/configs/figure5/medium/lstm/seed_16/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 1600000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_17/final.yaml b/configs/figure5/medium/lstm/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f7abb125628529e521b43c5189dc2931e6b38806 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_17/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 1700000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_18/final.yaml b/configs/figure5/medium/lstm/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..de9befa02d5c51683aab9a62693896d2046d5b4e --- /dev/null +++ b/configs/figure5/medium/lstm/seed_18/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 1800000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_19/final.yaml b/configs/figure5/medium/lstm/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..cc48af3482b6fe9e87d0b8657085b8fb1f0cb980 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_19/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 1900000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_2/final.yaml b/configs/figure5/medium/lstm/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5161a7af5cd01eda9236cdfba1e6310c30bce115 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_2/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 200000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_3/final.yaml b/configs/figure5/medium/lstm/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c0e38103ddd1b990d0179765d8fcaeff552996f9 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_3/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 300000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_4/final.yaml b/configs/figure5/medium/lstm/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4f488732deebb8c24581f834ad5e65132c859da3 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_4/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 400000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_5/final.yaml b/configs/figure5/medium/lstm/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..570df41adc73ff81be14829b5d922ce091b9cd6b --- /dev/null +++ b/configs/figure5/medium/lstm/seed_5/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 500000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_6/final.yaml b/configs/figure5/medium/lstm/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..00b2c2962e9cef837b3c534b744aeb07efa1b549 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_6/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 600000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_7/final.yaml b/configs/figure5/medium/lstm/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1261103f88a0da89ac44b35c7ff8e30e7ebde628 --- /dev/null +++ b/configs/figure5/medium/lstm/seed_7/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 700000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_8/final.yaml b/configs/figure5/medium/lstm/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2d76ae5fdeeba65a77b8ad6d54612fa91cea6aee --- /dev/null +++ b/configs/figure5/medium/lstm/seed_8/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 800000 + evaluation: 9300000 diff --git a/configs/figure5/medium/lstm/seed_9/final.yaml b/configs/figure5/medium/lstm/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1887d110ebb9c2c00692b37ddb691daf3c1aaf3b --- /dev/null +++ b/configs/figure5/medium/lstm/seed_9/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: lstm + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 900000 + evaluation: 9300000 diff --git a/configs/figure5/medium/mlp/seed_0/final.yaml b/configs/figure5/medium/mlp/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c6d20c491a13239eaec385a6c63db76d226d48d1 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_0/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 0 + evaluation: 120000 diff --git a/configs/figure5/medium/mlp/seed_1/final.yaml b/configs/figure5/medium/mlp/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e2ad9c5b676bca5c55fc6856c64b823ce11fba0f --- /dev/null +++ b/configs/figure5/medium/mlp/seed_1/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 0 + evaluation: 120000 diff --git a/configs/figure5/medium/mlp/seed_10/final.yaml b/configs/figure5/medium/mlp/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..340a0f87a814cd65477cfdcd76ba98118e289004 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_10/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_11/final.yaml b/configs/figure5/medium/mlp/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..277f4fbc2373c3848e6e39bd167b46e769b67ce8 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_11/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_12/final.yaml b/configs/figure5/medium/mlp/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dd91206331a4293297f5193d9b795799e1bc09f3 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_12/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_13/final.yaml b/configs/figure5/medium/mlp/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e924f3a25365a7bb5c942c6c9ca9160943821118 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_13/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_14/final.yaml b/configs/figure5/medium/mlp/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3af3fe96b19a5b82f2bc0bf21d8d1ed8e850d1b8 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_14/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_15/final.yaml b/configs/figure5/medium/mlp/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6256246d922ee70ad5d446e386b5fbbec5f7efc0 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_15/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_16/final.yaml b/configs/figure5/medium/mlp/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..760cd24b4b830a275d48b9db53d39ef820309502 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_16/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_17/final.yaml b/configs/figure5/medium/mlp/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ccf109778df849c432536cd6dcc3036100b52957 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_17/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_18/final.yaml b/configs/figure5/medium/mlp/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fe933c7245e63d4ec9da26c871b257ce7bf1ec88 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_18/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_19/final.yaml b/configs/figure5/medium/mlp/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..665e0a8d7d9f4ee7e1ac554486985e90ab005dca --- /dev/null +++ b/configs/figure5/medium/mlp/seed_19/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_2/final.yaml b/configs/figure5/medium/mlp/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c1f59380713c411fdf028dbbbc3802fb9a8c267f --- /dev/null +++ b/configs/figure5/medium/mlp/seed_2/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 0 + evaluation: 120000 diff --git a/configs/figure5/medium/mlp/seed_3/final.yaml b/configs/figure5/medium/mlp/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c1b6909901147a30c85eab5e8008219763d40512 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_3/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 0 + evaluation: 120000 diff --git a/configs/figure5/medium/mlp/seed_4/final.yaml b/configs/figure5/medium/mlp/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..995036f302dc5b86bd2e6c93164575aad2305f98 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_4/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 0 + evaluation: 120000 diff --git a/configs/figure5/medium/mlp/seed_5/final.yaml b/configs/figure5/medium/mlp/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2cd8c906e677fd530350a30d1b858ac66245992b --- /dev/null +++ b/configs/figure5/medium/mlp/seed_5/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_6/final.yaml b/configs/figure5/medium/mlp/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..24051f1716cfc3526ff9117f9817166d4eaf4a87 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_6/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_7/final.yaml b/configs/figure5/medium/mlp/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dedf4dc9083916088e2d22f09b52a910a5ab66a3 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_7/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_8/final.yaml b/configs/figure5/medium/mlp/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..de1bdd219b760c9f9839c28521875f5245f4f0b3 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_8/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/mlp/seed_9/final.yaml b/configs/figure5/medium/mlp/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d1996b65d5f7de1942fc187285ced8998137a8e4 --- /dev/null +++ b/configs/figure5/medium/mlp/seed_9/final.yaml @@ -0,0 +1,97 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: mlp + actor_kwargs: + hidden: + - 256 + - 256 + activation: relu + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 8 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.8 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 0 + evaluation: 9310000 diff --git a/configs/figure5/medium/reservoir/seed_0/final.yaml b/configs/figure5/medium/reservoir/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..399b7413eba272674b0dab6ebbd81922520752c8 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_0/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 0 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_1/final.yaml b/configs/figure5/medium/reservoir/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..147b11acd84627ef6c2ca2bda95fa6cfbc813705 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_1/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 100000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_10/final.yaml b/configs/figure5/medium/reservoir/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fa5437dd57e55d94344ef9915cec982680c7b42b --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_10/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 1000000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_11/final.yaml b/configs/figure5/medium/reservoir/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3a9c393d402111bb2a03373f3dafd17e262b1dc9 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_11/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 1100000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_12/final.yaml b/configs/figure5/medium/reservoir/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ae2a8e0b5ddeea3f638239fafa5b0dcaec4cb38f --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_12/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 1200000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_13/final.yaml b/configs/figure5/medium/reservoir/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5780213307907dae68b6d19aba4d89c5d9c92ded --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_13/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 1300000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_14/final.yaml b/configs/figure5/medium/reservoir/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1787b7965414a15f82f06cf45f940ff1e5539acd --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_14/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 1400000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_15/final.yaml b/configs/figure5/medium/reservoir/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1ba8cd7448862f1a98cbda8197f74caa0e0a6e96 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_15/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 1500000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_16/final.yaml b/configs/figure5/medium/reservoir/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2e4c185310f60c3d54d0178734a436b05bdb8230 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_16/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 1600000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_17/final.yaml b/configs/figure5/medium/reservoir/seed_17/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6f54b191397ecf989ec1620fce802728c8773217 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_17/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 17 +seed_roles: + optimization: 17 + sampling: 17 + collection: 1700000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_18/final.yaml b/configs/figure5/medium/reservoir/seed_18/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6de683f7248491f475c9f4c72771150980aa8f61 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_18/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 18 +seed_roles: + optimization: 18 + sampling: 18 + collection: 1800000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_19/final.yaml b/configs/figure5/medium/reservoir/seed_19/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5092fd58a691a33a520c14de07be6a938923309d --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_19/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 19 +seed_roles: + optimization: 19 + sampling: 19 + collection: 1900000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_2/final.yaml b/configs/figure5/medium/reservoir/seed_2/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..220a3ba248999d1f0acea1b61fc84b282fea3f22 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_2/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 2 +seed_roles: + optimization: 2 + sampling: 2 + collection: 200000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_3/final.yaml b/configs/figure5/medium/reservoir/seed_3/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..896b7b1a3f184c852e168e2bbcd1c060925e0427 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_3/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 3 +seed_roles: + optimization: 3 + sampling: 3 + collection: 300000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_4/final.yaml b/configs/figure5/medium/reservoir/seed_4/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6cbf08a06f0550b185639061e3bb5aecf9e7fca6 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_4/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 4 +seed_roles: + optimization: 4 + sampling: 4 + collection: 400000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_5/final.yaml b/configs/figure5/medium/reservoir/seed_5/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..aa5e25a39324733034da5d99a7168e7384ee575a --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_5/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 5 +seed_roles: + optimization: 5 + sampling: 5 + collection: 500000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_6/final.yaml b/configs/figure5/medium/reservoir/seed_6/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c8964eb19d4150db2c2df0f2ac78c309b0136f7e --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_6/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 6 +seed_roles: + optimization: 6 + sampling: 6 + collection: 600000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_7/final.yaml b/configs/figure5/medium/reservoir/seed_7/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f24eba6ebb79b87916341c18dc56e28df2581e0e --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_7/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 7 +seed_roles: + optimization: 7 + sampling: 7 + collection: 700000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_8/final.yaml b/configs/figure5/medium/reservoir/seed_8/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a460310d608257e545ce387d492341faeb50b6b4 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_8/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 8 +seed_roles: + optimization: 8 + sampling: 8 + collection: 800000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir/seed_9/final.yaml b/configs/figure5/medium/reservoir/seed_9/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..10a86e30524581edd36940f4b839e44a2091ea73 --- /dev/null +++ b/configs/figure5/medium/reservoir/seed_9/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 9 +seed_roles: + optimization: 9 + sampling: 9 + collection: 900000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir_ro/seed_0/final.yaml b/configs/figure5/medium/reservoir_ro/seed_0/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..186fcd0620460d460ef3282eb6ac6c4a85534458 --- /dev/null +++ b/configs/figure5/medium/reservoir_ro/seed_0/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 0 +seed_roles: + optimization: 0 + sampling: 0 + collection: 0 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir_ro/seed_1/final.yaml b/configs/figure5/medium/reservoir_ro/seed_1/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2b7530ccd0bfbbad6a794929fbf0b9c44198b5c6 --- /dev/null +++ b/configs/figure5/medium/reservoir_ro/seed_1/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 1 +seed_roles: + optimization: 1 + sampling: 1 + collection: 100000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir_ro/seed_10/final.yaml b/configs/figure5/medium/reservoir_ro/seed_10/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4b4f95348c312f09528fdb9cc9b942cecedd07bd --- /dev/null +++ b/configs/figure5/medium/reservoir_ro/seed_10/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 10 +seed_roles: + optimization: 10 + sampling: 10 + collection: 1000000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir_ro/seed_11/final.yaml b/configs/figure5/medium/reservoir_ro/seed_11/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c2890662b87ab88032f1409951998230b23fc26f --- /dev/null +++ b/configs/figure5/medium/reservoir_ro/seed_11/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 11 +seed_roles: + optimization: 11 + sampling: 11 + collection: 1100000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir_ro/seed_12/final.yaml b/configs/figure5/medium/reservoir_ro/seed_12/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2d346db7bccc15662ba24c10d2e46db09655c386 --- /dev/null +++ b/configs/figure5/medium/reservoir_ro/seed_12/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 12 +seed_roles: + optimization: 12 + sampling: 12 + collection: 1200000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir_ro/seed_13/final.yaml b/configs/figure5/medium/reservoir_ro/seed_13/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f5656fac94c3fa90bd7782a4ea993cf8318f541f --- /dev/null +++ b/configs/figure5/medium/reservoir_ro/seed_13/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 13 +seed_roles: + optimization: 13 + sampling: 13 + collection: 1300000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir_ro/seed_14/final.yaml b/configs/figure5/medium/reservoir_ro/seed_14/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ab93e9bb9682b2f7e08aa212c16a855859be3443 --- /dev/null +++ b/configs/figure5/medium/reservoir_ro/seed_14/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 14 +seed_roles: + optimization: 14 + sampling: 14 + collection: 1400000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir_ro/seed_15/final.yaml b/configs/figure5/medium/reservoir_ro/seed_15/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..73eb3645a0f2e5bb4c163edf6f67fae3c0c5c20a --- /dev/null +++ b/configs/figure5/medium/reservoir_ro/seed_15/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 15 +seed_roles: + optimization: 15 + sampling: 15 + collection: 1500000 + evaluation: 9300000 diff --git a/configs/figure5/medium/reservoir_ro/seed_16/final.yaml b/configs/figure5/medium/reservoir_ro/seed_16/final.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c1cbfc268de748acb64944af5554451330a2b5b5 --- /dev/null +++ b/configs/figure5/medium/reservoir_ro/seed_16/final.yaml @@ -0,0 +1,93 @@ +algorithm: antmaze_teacher_distillation +observation_clip: 30.0 +policy: + name: reservoir_ro + actor_kwargs: + goal_router_hidden: + - 256 + - 256 + goal_router_codebook: + - - -6.0 + - 10.0 + - - -10.0 + - 6.0 + - - -6.0 + - 6.0 + - - 2.0 + - 6.0 + - - 6.0 + - 6.0 + - - -6.0 + - 2.0 + - - -2.0 + - 2.0 + - - 2.0 + - 2.0 + - - -10.0 + - -2.0 + - - -6.0 + - -2.0 + - - 2.0 + - -2.0 + - - 6.0 + - -2.0 + - - 10.0 + - -2.0 + - - 10.0 + - -6.0 + - - 10.0 + - -10.0 + goal_router_codebook_latch_radius: 0.45 + goal_router_codebook_active_pairs: + - - - 6.0 + - -10.0 + - - -10.0 + - -6.0 + - - - 6.0 + - -10.0 + - - -10.0 + - 10.0 + - - - -10.0 + - -6.0 + - - 10.0 + - 6.0 + goal_router_codebook_pair_match_radius: 2.0 + value_hidden: + - 256 + - 256 + value_activation: relu +env: + class_path: larva.ant_transfer.env.AntMazeTransferEnv + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +resolved_env: + env_id: AntMaze_Medium_Diverse_GR-v5 + goal_feature_mode: world + continuing_task: false +imitation: + demonstration_steps: 60000 + dagger_steps: 30000 + dagger_iterations: 0 + updates_per_stage: 5000 + batch_size: 64 + burn_in: 16 + learning_rate: 0.0001 + gradient_clip: 1.0 + actor_train_mode: base_all + goal_input_features: 4 + loss_space: pre_tanh + sampling_mode: episode_uniform + checkpoint_interval_updates: 500 +navigation: + waypoint_planner: official_grid_waypoints + student_goal: final + evaluation_waypoint_planner: none +evaluation: + action_smoothing: 0.95 +seed: 16 +seed_roles: + optimization: 16 + sampling: 16 + collection: 1600000 + evaluation: 9300000