Archive paper Figures 2–4 and S3–S6 for private review (part 2)

configs/S4/larva/seed_4.yaml

@@ -0,0 +1,107 @@
+connectome:
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_larva6_20260928/graphs/larva_row_l1.pkl
+  frozen_graph_sha256: 255dfe9d4dedba1b08d02bf4140a650de4cd80d6778b2e14d487fd1be2be6c76
+  global_gain: 1.0
+  kind: original
+  normalization: row_l1
+  sensory_type: all
+  species: larva
+continuous_gait_experiment:
+  action_dimension_matched: true
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/configs/adult_seed4.yaml
+  cpg_freedom: duty_warping_plus_phase_family
+  primary_checkpoint: fixed_budget_final
+  target_sampling: uniform_episode
+  training_unit: one fresh Larva seed, matching Adult training protocol
+env:
+  action_mode: command2
+  command_adapter_config:
+    base_frequency_hz: 12.0
+    continuous_gait: true
+    fast_duty_factor: 0.58
+    fast_ipsilateral_lag_cycles: 0.5
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    gait_smoothing_alpha: 0.005
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    slow_duty_factor: 0.72
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    step_bounds:
+    - 0.4
+    - 2.0
+    swing_duration_s: 0.035
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+  speed_target_sampling: uniform
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  task: speed_control
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_larva6_20260928/initialization/critic_seed4.pt
+interface_hierarchy:
+  batch: EXP04_continuous_speed_larva6_20260928
+  connectome: larva
+  interface: command2
+  label: larva
+  smoke: false
+  task: speed_control
+  train_seed: 4
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  environment_seed: 14000
+  gamma: 0.99
+  initial_hidden_mode: zero
+  initialization_seed: 4
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  minibatch_seed: 30004
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  policy_sampling_seed: 20004
+  rollout_heartbeat: 32
+  separate_rng_streams: true
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+resume_policy: restart_only
+seed: 4
+vec_env: subproc

configs/S4/larva/seed_5.yaml

@@ -0,0 +1,107 @@
+connectome:
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_larva6_20260928/graphs/larva_row_l1.pkl
+  frozen_graph_sha256: 255dfe9d4dedba1b08d02bf4140a650de4cd80d6778b2e14d487fd1be2be6c76
+  global_gain: 1.0
+  kind: original
+  normalization: row_l1
+  sensory_type: all
+  species: larva
+continuous_gait_experiment:
+  action_dimension_matched: true
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/configs/adult_seed5.yaml
+  cpg_freedom: duty_warping_plus_phase_family
+  primary_checkpoint: fixed_budget_final
+  target_sampling: uniform_episode
+  training_unit: one fresh Larva seed, matching Adult training protocol
+env:
+  action_mode: command2
+  command_adapter_config:
+    base_frequency_hz: 12.0
+    continuous_gait: true
+    fast_duty_factor: 0.58
+    fast_ipsilateral_lag_cycles: 0.5
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    gait_smoothing_alpha: 0.005
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    slow_duty_factor: 0.72
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    step_bounds:
+    - 0.4
+    - 2.0
+    swing_duration_s: 0.035
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+  speed_target_sampling: uniform
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  task: speed_control
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_larva6_20260928/initialization/critic_seed5.pt
+interface_hierarchy:
+  batch: EXP04_continuous_speed_larva6_20260928
+  connectome: larva
+  interface: command2
+  label: larva
+  smoke: false
+  task: speed_control
+  train_seed: 5
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  environment_seed: 15000
+  gamma: 0.99
+  initial_hidden_mode: zero
+  initialization_seed: 5
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  minibatch_seed: 30005
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  policy_sampling_seed: 20005
+  rollout_heartbeat: 32
+  separate_rng_streams: true
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+resume_policy: restart_only
+seed: 5
+vec_env: subproc

configs/S5/banc_descending/seed_0.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_descending
+  interface: command2
+  task: speed_control
+  train_seed: 0
+  smoke: false
+  label: banc_descending
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - descending
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_descending.pkl
+  frozen_graph_sha256: 86468ba0ad24d42cb8507741c05830da720573cf3df188574d6ffa1309aad915
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 0
+  environment_seed: 10000
+  policy_sampling_seed: 20000
+  minibatch_seed: 30000
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_descending_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 0
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed0.pt
+resume_policy: restart_only

configs/S5/banc_descending/seed_1.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_descending
+  interface: command2
+  task: speed_control
+  train_seed: 1
+  smoke: false
+  label: banc_descending
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - descending
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_descending.pkl
+  frozen_graph_sha256: 86468ba0ad24d42cb8507741c05830da720573cf3df188574d6ffa1309aad915
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 1
+  environment_seed: 11000
+  policy_sampling_seed: 20001
+  minibatch_seed: 30001
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_descending_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 1
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed1.pt
+resume_policy: restart_only

configs/S5/banc_descending/seed_2.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_descending
+  interface: command2
+  task: speed_control
+  train_seed: 2
+  smoke: false
+  label: banc_descending
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - descending
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_descending.pkl
+  frozen_graph_sha256: 86468ba0ad24d42cb8507741c05830da720573cf3df188574d6ffa1309aad915
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 2
+  environment_seed: 12000
+  policy_sampling_seed: 20002
+  minibatch_seed: 30002
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_descending_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 2
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed2.pt
+resume_policy: restart_only

configs/S5/banc_descending/seed_3.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_descending
+  interface: command2
+  task: speed_control
+  train_seed: 3
+  smoke: false
+  label: banc_descending
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - descending
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_descending.pkl
+  frozen_graph_sha256: 86468ba0ad24d42cb8507741c05830da720573cf3df188574d6ffa1309aad915
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 3
+  environment_seed: 13000
+  policy_sampling_seed: 20003
+  minibatch_seed: 30003
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_descending_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 3
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed3.pt
+resume_policy: restart_only

configs/S5/banc_descending/seed_4.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_descending
+  interface: command2
+  task: speed_control
+  train_seed: 4
+  smoke: false
+  label: banc_descending
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - descending
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_descending.pkl
+  frozen_graph_sha256: 86468ba0ad24d42cb8507741c05830da720573cf3df188574d6ffa1309aad915
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 4
+  environment_seed: 14000
+  policy_sampling_seed: 20004
+  minibatch_seed: 30004
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_descending_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 4
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed4.pt
+resume_policy: restart_only

configs/S5/banc_descending/seed_5.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_descending
+  interface: command2
+  task: speed_control
+  train_seed: 5
+  smoke: false
+  label: banc_descending
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - descending
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_descending.pkl
+  frozen_graph_sha256: 86468ba0ad24d42cb8507741c05830da720573cf3df188574d6ffa1309aad915
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 5
+  environment_seed: 15000
+  policy_sampling_seed: 20005
+  minibatch_seed: 30005
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_descending_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 5
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed5.pt
+resume_policy: restart_only

configs/S5/banc_motor/seed_0.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_motor
+  interface: command2
+  task: speed_control
+  train_seed: 0
+  smoke: false
+  label: banc_motor
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - motor
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_motor.pkl
+  frozen_graph_sha256: dde3a9e09030e301719ddf71c3eacc08bdaaeda009192310bcf0c39980859b59
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 0
+  environment_seed: 10000
+  policy_sampling_seed: 20000
+  minibatch_seed: 30000
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_motor_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 0
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed0.pt
+resume_policy: restart_only

configs/S5/banc_motor/seed_1.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_motor
+  interface: command2
+  task: speed_control
+  train_seed: 1
+  smoke: false
+  label: banc_motor
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - motor
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_motor.pkl
+  frozen_graph_sha256: dde3a9e09030e301719ddf71c3eacc08bdaaeda009192310bcf0c39980859b59
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 1
+  environment_seed: 11000
+  policy_sampling_seed: 20001
+  minibatch_seed: 30001
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_motor_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 1
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed1.pt
+resume_policy: restart_only

configs/S5/banc_motor/seed_2.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_motor
+  interface: command2
+  task: speed_control
+  train_seed: 2
+  smoke: false
+  label: banc_motor
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - motor
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_motor.pkl
+  frozen_graph_sha256: dde3a9e09030e301719ddf71c3eacc08bdaaeda009192310bcf0c39980859b59
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 2
+  environment_seed: 12000
+  policy_sampling_seed: 20002
+  minibatch_seed: 30002
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_motor_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 2
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed2.pt
+resume_policy: restart_only

configs/S5/banc_motor/seed_3.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_motor
+  interface: command2
+  task: speed_control
+  train_seed: 3
+  smoke: false
+  label: banc_motor
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - motor
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_motor.pkl
+  frozen_graph_sha256: dde3a9e09030e301719ddf71c3eacc08bdaaeda009192310bcf0c39980859b59
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 3
+  environment_seed: 13000
+  policy_sampling_seed: 20003
+  minibatch_seed: 30003
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_motor_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 3
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed3.pt
+resume_policy: restart_only

configs/S5/banc_motor/seed_4.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_motor
+  interface: command2
+  task: speed_control
+  train_seed: 4
+  smoke: false
+  label: banc_motor
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - motor
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_motor.pkl
+  frozen_graph_sha256: dde3a9e09030e301719ddf71c3eacc08bdaaeda009192310bcf0c39980859b59
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 4
+  environment_seed: 14000
+  policy_sampling_seed: 20004
+  minibatch_seed: 30004
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_motor_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 4
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed4.pt
+resume_policy: restart_only

configs/S5/banc_motor/seed_5.yaml

@@ -0,0 +1,120 @@
+interface_hierarchy:
+  batch: EXP04_continuous_speed_matched6_20260921
+  connectome: banc_motor
+  interface: command2
+  task: speed_control
+  train_seed: 5
+  smoke: false
+  label: banc_motor
+vec_env: subproc
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - motor
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+  frozen_graph_path: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/graphs/banc_motor.pkl
+  frozen_graph_sha256: dde3a9e09030e301719ddf71c3eacc08bdaaeda009192310bcf0c39980859b59
+env:
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+  command_adapter_config:
+    frequency_bounds:
+    - 0.25
+    - 2.0
+    k_drive: 0.75
+    k_drive_step: 1.0
+    k_step: 0.75
+    step_bounds:
+    - 0.4
+    - 2.0
+    continuous_gait: true
+    slow_duty_factor: 0.72
+    fast_duty_factor: 0.58
+    slow_ipsilateral_lag_cycles: 0.3333333333333333
+    fast_ipsilateral_lag_cycles: 0.5
+    swing_duration_s: 0.035
+    base_frequency_hz: 12.0
+    gait_smoothing_alpha: 0.005
+  speed_filter_alpha: 0.25
+  speed_heading_rate_weight: 0.25
+  speed_target_scale_mm_s: 20.0
+  speed_targets_mm_s:
+  - 8.0
+  - 16.0
+  action_mode: command2
+  task: speed_control
+  speed_target_sampling: uniform
+  speed_target_range_mm_s:
+  - 8.0
+  - 16.0
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 32
+  seq_len: 32
+  total_timesteps: 499200
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+  separate_rng_streams: true
+  initialization_seed: 5
+  environment_seed: 15000
+  policy_sampling_seed: 20005
+  minibatch_seed: 30005
+  initial_hidden_mode: zero
+continuous_gait_experiment:
+  baseline: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed/configs/speed_continuous_gait_banc_motor_command2_seed0.yaml
+  target_sampling: uniform_episode
+  cpg_freedom: duty_warping_plus_phase_family
+  action_dimension_matched: true
+  primary_checkpoint: fixed_budget_final
+  training_unit: one complete fresh train seed; ER also independent graph seed
+seed: 5
+initialization:
+  critic_state_from: /data2/ysy/github/flygym/outputs/nmi_unified/experiments/EXP04_continuous_speed_matched6_20260921/initialization/critic_seed5.pt
+resume_policy: restart_only

configs/S6/banc_descending/seed_0.yaml

@@ -0,0 +1,79 @@
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - descending
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+env:
+  action_mode: command2
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  perturbation:
+    direction: lateral
+    duration_s: 0.04
+    force: 10.0
+    start_time_s: 0.75
+    type: side_push
+  task: heading_hold
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+interface_hierarchy:
+  batch: interface_hierarchy_original
+  connectome: banc_descending_original
+  interface: command2
+  smoke: false
+  task: heading_hold
+  train_seed: 0
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 8
+  seq_len: 32
+  total_timesteps: 500000
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+seed: 0
+vec_env: subproc

configs/S6/banc_descending/seed_1.yaml

@@ -0,0 +1,79 @@
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - descending
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+env:
+  action_mode: command2
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  perturbation:
+    direction: lateral
+    duration_s: 0.04
+    force: 10.0
+    start_time_s: 0.75
+    type: side_push
+  task: heading_hold
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+interface_hierarchy:
+  batch: interface_hierarchy_original
+  connectome: banc_descending_original
+  interface: command2
+  smoke: false
+  task: heading_hold
+  train_seed: 1
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 8
+  seq_len: 32
+  total_timesteps: 500000
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+seed: 1
+vec_env: subproc

configs/S6/banc_descending/seed_2.yaml

@@ -0,0 +1,79 @@
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - descending
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+env:
+  action_mode: command2
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  perturbation:
+    direction: lateral
+    duration_s: 0.04
+    force: 10.0
+    start_time_s: 0.75
+    type: side_push
+  task: heading_hold
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+interface_hierarchy:
+  batch: interface_hierarchy_original
+  connectome: banc_descending_original
+  interface: command2
+  smoke: false
+  task: heading_hold
+  train_seed: 2
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 8
+  seq_len: 32
+  total_timesteps: 500000
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+seed: 2
+vec_env: subproc

configs/S6/banc_motor/seed_0.yaml

@@ -0,0 +1,79 @@
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - motor
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+env:
+  action_mode: command2
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  perturbation:
+    direction: lateral
+    duration_s: 0.04
+    force: 10.0
+    start_time_s: 0.75
+    type: side_push
+  task: heading_hold
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+interface_hierarchy:
+  batch: interface_hierarchy_original
+  connectome: banc_original
+  interface: command2
+  smoke: false
+  task: heading_hold
+  train_seed: 0
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 8
+  seq_len: 32
+  total_timesteps: 500000
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+seed: 0
+vec_env: subproc

configs/S6/banc_motor/seed_1.yaml

@@ -0,0 +1,79 @@
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - motor
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+env:
+  action_mode: command2
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  perturbation:
+    direction: lateral
+    duration_s: 0.04
+    force: 10.0
+    start_time_s: 0.75
+    type: side_push
+  task: heading_hold
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+interface_hierarchy:
+  batch: interface_hierarchy_original
+  connectome: banc_original
+  interface: command2
+  smoke: false
+  task: heading_hold
+  train_seed: 1
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 8
+  seq_len: 32
+  total_timesteps: 500000
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+seed: 1
+vec_env: subproc

configs/S6/banc_motor/seed_2.yaml

@@ -0,0 +1,79 @@
+connectome:
+  banc_dir: larva/data/banc
+  chunksize: 250000
+  global_gain: 1.0
+  kind: original
+  min_syn_count: 1
+  normalization: row_l1
+  output_selector:
+    field: super_class
+    include:
+    - motor
+  polarity:
+    unknown_action: zero
+  randomize_seed: 0
+  sensory_selector:
+    field: flow
+    include:
+    - afferent
+  species: banc
+env:
+  action_mode: command2
+  decision_interval: 100
+  flip_penalty: 1000.0
+  max_correction: 10.0
+  max_time: 1.5
+  observation_mode: proprio_task
+  perturbation:
+    direction: lateral
+    duration_s: 0.04
+    force: 10.0
+    start_time_s: 0.75
+    type: side_push
+  task: heading_hold
+  terrain_seeds:
+  - 0
+  - 1
+  - 2
+  - 3
+  terrains:
+  - flat
+  timestep: 0.0001
+interface_hierarchy:
+  batch: interface_hierarchy_original
+  connectome: banc_original
+  interface: command2
+  smoke: false
+  task: heading_hold
+  train_seed: 2
+policy:
+  actor_kwargs:
+    dt: 0.01
+    input_target: sensory
+    mode: rate
+    tau_init: 0.02
+    use_lora: false
+  log_std_init: -0.5
+  name: connectome_cfc
+  value_hidden:
+  - 128
+  - 128
+ppo:
+  clip_range: 0.2
+  ent_coef: 0.0
+  gamma: 0.99
+  lam: 0.95
+  lr: 0.0001
+  max_grad_norm: 0.5
+  n_envs: 16
+  n_epochs: 2
+  n_minibatches: 5
+  n_steps: 160
+  normalize_reward: true
+  rollout_heartbeat: 8
+  seq_len: 32
+  total_timesteps: 500000
+  vf_clip_range: 0.2
+  vf_coef: 0.5
+seed: 2
+vec_env: subproc

configs/figure2/connectome/seed_2011.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2011,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2011,
+    "separate_rng_streams": true,
+    "initialization_seed": 2011,
+    "environment_seed": 5201100,
+    "policy_sampling_seed": 102011,
+    "minibatch_seed": 302011,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "663b9123cb0942fd2bbd98a8da7aaa14968116780a24042a28f6cb213dba4302",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2012.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2012,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2012,
+    "separate_rng_streams": true,
+    "initialization_seed": 2012,
+    "environment_seed": 5201200,
+    "policy_sampling_seed": 102012,
+    "minibatch_seed": 302012,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "27e7d4589ae2a02d49168ee3130056bc39e7d5acf80c93c5c78bec08901a3270",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2013.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2013,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2013,
+    "separate_rng_streams": true,
+    "initialization_seed": 2013,
+    "environment_seed": 5201300,
+    "policy_sampling_seed": 102013,
+    "minibatch_seed": 302013,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "e005816168be1f9a68a4e86d709e31c4746a591ad27106f580f1a8ac8146188d",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2014.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2014,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2014,
+    "separate_rng_streams": true,
+    "initialization_seed": 2014,
+    "environment_seed": 5201400,
+    "policy_sampling_seed": 102014,
+    "minibatch_seed": 302014,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "1d01c86fdf54cee2689893193bad53dccd5ddc7662e3041d705c57aae05db9cf",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2015.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2015,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2015,
+    "separate_rng_streams": true,
+    "initialization_seed": 2015,
+    "environment_seed": 5201500,
+    "policy_sampling_seed": 102015,
+    "minibatch_seed": 302015,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "0f0f10c0bb695923b99f909614ff2b05cf2d426a5fc8489cae9cba88bc4128a0",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2016.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2016,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2016,
+    "separate_rng_streams": true,
+    "initialization_seed": 2016,
+    "environment_seed": 5201600,
+    "policy_sampling_seed": 102016,
+    "minibatch_seed": 302016,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "e7ac33a797fccb430fc12d7dc419a307a4bbe8148107f9394d6c324e4502e3cb",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2017.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2017,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2017,
+    "separate_rng_streams": true,
+    "initialization_seed": 2017,
+    "environment_seed": 5201700,
+    "policy_sampling_seed": 102017,
+    "minibatch_seed": 302017,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "af0e370c98101dae66fc3ae5f66047d573fb767917e2d74319069cc75f82ca6b",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2018.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2018,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2018,
+    "separate_rng_streams": true,
+    "initialization_seed": 2018,
+    "environment_seed": 5201800,
+    "policy_sampling_seed": 102018,
+    "minibatch_seed": 302018,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "473c2abd5dbf0604b19594a05671430bbbea667f02e8df4e3c292cd0ea02f089",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2019.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2019,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2019,
+    "separate_rng_streams": true,
+    "initialization_seed": 2019,
+    "environment_seed": 5201900,
+    "policy_sampling_seed": 102019,
+    "minibatch_seed": 302019,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "987b4cf36337646a2ede352a26900de76ac49af659af6519caa7809c692d1c4f",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2020.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2020,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2020,
+    "separate_rng_streams": true,
+    "initialization_seed": 2020,
+    "environment_seed": 5202000,
+    "policy_sampling_seed": 102020,
+    "minibatch_seed": 302020,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "9d7ce49cb0c96ed0ec9af1e5b73c9f93c7ee35a74182a249c5fdcc14375058ca",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2021.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2021,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2021,
+    "separate_rng_streams": true,
+    "initialization_seed": 2021,
+    "environment_seed": 5202100,
+    "policy_sampling_seed": 102021,
+    "minibatch_seed": 302021,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "a0121d8d6fb53980e9a2e0056a70efa9405e7894de5b84d34340b03b9b936cfe",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2022.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2022,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2022,
+    "separate_rng_streams": true,
+    "initialization_seed": 2022,
+    "environment_seed": 5202200,
+    "policy_sampling_seed": 102022,
+    "minibatch_seed": 302022,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "5cf8fe449402a9217e565ce53111bc58b2e6a2f3a8df2ebe904ec9afbd81665e",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2023.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2023,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2023,
+    "separate_rng_streams": true,
+    "initialization_seed": 2023,
+    "environment_seed": 5202300,
+    "policy_sampling_seed": 102023,
+    "minibatch_seed": 302023,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "40b3c46e8e20197007f9d394bb9a282d707df23f73e8aede75991084df8dd5e2",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2024.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2024,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2024,
+    "separate_rng_streams": true,
+    "initialization_seed": 2024,
+    "environment_seed": 5202400,
+    "policy_sampling_seed": 102024,
+    "minibatch_seed": 302024,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "d3d564c0b8901fee3bdfccdce8fb678b4bda7ee8f7c9454ca2601eee8d8a8048",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2025.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2025,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2025,
+    "separate_rng_streams": true,
+    "initialization_seed": 2025,
+    "environment_seed": 5202500,
+    "policy_sampling_seed": 102025,
+    "minibatch_seed": 302025,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "8e8ef57e505352aadb496afc69ff08ee15caf9e52cb82f6b7773aeff701e4617",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2026.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2026,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2026,
+    "separate_rng_streams": true,
+    "initialization_seed": 2026,
+    "environment_seed": 5202600,
+    "policy_sampling_seed": 102026,
+    "minibatch_seed": 302026,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "014bc75abbb7047ae99c521fd34b3179c0a5462b31f02ec6a15a06da44e9459f",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2027.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2027,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2027,
+    "separate_rng_streams": true,
+    "initialization_seed": 2027,
+    "environment_seed": 5202700,
+    "policy_sampling_seed": 102027,
+    "minibatch_seed": 302027,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "4c4ad13c44d5d4df2a7347d2f954755c3f76955e122bd735a2967543f74a374f",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2028.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2028,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2028,
+    "separate_rng_streams": true,
+    "initialization_seed": 2028,
+    "environment_seed": 5202800,
+    "policy_sampling_seed": 102028,
+    "minibatch_seed": 302028,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "fee737e447bd4ab6542e53d21fe271887f6a88090634decd0b8925d253a15d4e",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2029.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2029,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2029,
+    "separate_rng_streams": true,
+    "initialization_seed": 2029,
+    "environment_seed": 5202900,
+    "policy_sampling_seed": 102029,
+    "minibatch_seed": 302029,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "125c5cbcab296c4cd98a6af032c440e8eb82a722c43a6165045e00c0f4b601a8",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/connectome/seed_2030.json

@@ -0,0 +1,61 @@
+{
+  "method": "connectome",
+  "seed": 2030,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2030,
+    "separate_rng_streams": true,
+    "initialization_seed": 2030,
+    "environment_seed": 5203000,
+    "policy_sampling_seed": 102030,
+    "minibatch_seed": 302030,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "8f953b7060aefdf6c736b37f2608446408210afc49072c99e0617134e0ccc0a3",
+  "graph_sha256": "547da3a28a9c7acb279cbc071f7582cbc45f3da3f07503e0d042980fb99eaa03",
+  "graph_buffers_sha256": "defe28edec2e000943bd7709e3439881e58a1e42bb4d9b870025e82826807ca7",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/er/seed_2011.json

@@ -0,0 +1,61 @@
+{
+  "method": "er",
+  "seed": 2011,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2011,
+    "separate_rng_streams": true,
+    "initialization_seed": 2011,
+    "environment_seed": 5201100,
+    "policy_sampling_seed": 102011,
+    "minibatch_seed": 302011,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "663b9123cb0942fd2bbd98a8da7aaa14968116780a24042a28f6cb213dba4302",
+  "graph_sha256": "af7bb7c42684e85f813d8fb53357cc5471a447f6b7088092822f4fc9ef4fb492",
+  "graph_buffers_sha256": "b42a1725f17c3740ce5a3b7db45688aca856797d05a1346a1917e7776d5080d2",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/er/seed_2012.json

@@ -0,0 +1,61 @@
+{
+  "method": "er",
+  "seed": 2012,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2012,
+    "separate_rng_streams": true,
+    "initialization_seed": 2012,
+    "environment_seed": 5201200,
+    "policy_sampling_seed": 102012,
+    "minibatch_seed": 302012,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "27e7d4589ae2a02d49168ee3130056bc39e7d5acf80c93c5c78bec08901a3270",
+  "graph_sha256": "1e31e4c7fb4a3d4e626e209f87c3bb84a5beae4b64c3b2db341769debe5c7cb0",
+  "graph_buffers_sha256": "61699a00c20da810d8be9c72eab5aabd5174da9e8a3a88e389fb50856e28ab4f",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/er/seed_2013.json

@@ -0,0 +1,61 @@
+{
+  "method": "er",
+  "seed": 2013,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2013,
+    "separate_rng_streams": true,
+    "initialization_seed": 2013,
+    "environment_seed": 5201300,
+    "policy_sampling_seed": 102013,
+    "minibatch_seed": 302013,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "e005816168be1f9a68a4e86d709e31c4746a591ad27106f580f1a8ac8146188d",
+  "graph_sha256": "8e512a5165514a154f3dc4e00b22584a515dabc02a84f6cadc53979aaf509ee1",
+  "graph_buffers_sha256": "501af264917a4ddefe0df64b2e2126ea0ac29385b218a3601bde830436cd990d",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/er/seed_2014.json

@@ -0,0 +1,61 @@
+{
+  "method": "er",
+  "seed": 2014,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2014,
+    "separate_rng_streams": true,
+    "initialization_seed": 2014,
+    "environment_seed": 5201400,
+    "policy_sampling_seed": 102014,
+    "minibatch_seed": 302014,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "1d01c86fdf54cee2689893193bad53dccd5ddc7662e3041d705c57aae05db9cf",
+  "graph_sha256": "721c7528d0969ad088566aa81a59658e5b0617ecb749a503ef14ebf686b259f4",
+  "graph_buffers_sha256": "8f061f242dab13ed8e9bfc61c7992dcfb102c563ce376c117e5be70675d948a9",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/er/seed_2015.json

@@ -0,0 +1,61 @@
+{
+  "method": "er",
+  "seed": 2015,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2015,
+    "separate_rng_streams": true,
+    "initialization_seed": 2015,
+    "environment_seed": 5201500,
+    "policy_sampling_seed": 102015,
+    "minibatch_seed": 302015,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "0f0f10c0bb695923b99f909614ff2b05cf2d426a5fc8489cae9cba88bc4128a0",
+  "graph_sha256": "0dd14788c19f7e4c007d2d0d900c499ffaf494747d15f6e4d5454c074aa7bff9",
+  "graph_buffers_sha256": "4677e7f045eb43036acd3cd54ccb2b4bb5a5137182168aee99260fe3276b63b2",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/er/seed_2016.json

@@ -0,0 +1,61 @@
+{
+  "method": "er",
+  "seed": 2016,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2016,
+    "separate_rng_streams": true,
+    "initialization_seed": 2016,
+    "environment_seed": 5201600,
+    "policy_sampling_seed": 102016,
+    "minibatch_seed": 302016,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "e7ac33a797fccb430fc12d7dc419a307a4bbe8148107f9394d6c324e4502e3cb",
+  "graph_sha256": "e03df681a124f88b006169542334357aa3bafaf6b570cb765467a9769c93121b",
+  "graph_buffers_sha256": "dea28f4414b185e078e4a39930daae96480b1b96f367e5aba3d50e167eea1c1d",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/er/seed_2017.json

@@ -0,0 +1,61 @@
+{
+  "method": "er",
+  "seed": 2017,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2017,
+    "separate_rng_streams": true,
+    "initialization_seed": 2017,
+    "environment_seed": 5201700,
+    "policy_sampling_seed": 102017,
+    "minibatch_seed": 302017,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "af0e370c98101dae66fc3ae5f66047d573fb767917e2d74319069cc75f82ca6b",
+  "graph_sha256": "8d89b097ea5df04a3cc90541f54f226b50ab64c457511a81c58bbe0eff0c432b",
+  "graph_buffers_sha256": "8fa55cb252067cda2d4212fb99731f15bd7817b762e2ea488f1158bb7a7e4d97",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/er/seed_2018.json

@@ -0,0 +1,61 @@
+{
+  "method": "er",
+  "seed": 2018,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2018,
+    "separate_rng_streams": true,
+    "initialization_seed": 2018,
+    "environment_seed": 5201800,
+    "policy_sampling_seed": 102018,
+    "minibatch_seed": 302018,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "473c2abd5dbf0604b19594a05671430bbbea667f02e8df4e3c292cd0ea02f089",
+  "graph_sha256": "028b562392a4bceae6a679aaeb7aae26241adbba8fb587b2305f6da6d4c24a62",
+  "graph_buffers_sha256": "dc95cc3fb1c3a7c7a97eb9a489ef4784605ad56f2fe028fcb9585cc36558c8ce",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/er/seed_2019.json

@@ -0,0 +1,61 @@
+{
+  "method": "er",
+  "seed": 2019,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2019,
+    "separate_rng_streams": true,
+    "initialization_seed": 2019,
+    "environment_seed": 5201900,
+    "policy_sampling_seed": 102019,
+    "minibatch_seed": 302019,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "987b4cf36337646a2ede352a26900de76ac49af659af6519caa7809c692d1c4f",
+  "graph_sha256": "a17874dac6acffbbd0e6713b12a795b58fb7f01ffe57ea36955d3a7a24877e11",
+  "graph_buffers_sha256": "1042e9bc3309815321fbb82770280469c6aea80f335f36b6c593601eb79151f0",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}

configs/figure2/er/seed_2020.json

@@ -0,0 +1,61 @@
+{
+  "method": "er",
+  "seed": 2020,
+  "config": {
+    "total_timesteps": 32768,
+    "n_envs": 8,
+    "n_steps": 32,
+    "seq_len": 8,
+    "n_epochs": 4,
+    "n_minibatches": 8,
+    "gamma": 0.9920319148370607,
+    "lam": 0.951229424500714,
+    "clip_range": 0.2,
+    "vf_clip_range": 0.2,
+    "lr": 0.0001,
+    "actor_lr": 0.0001,
+    "critic_lr": 0.0001,
+    "log_std_lr": 0.001,
+    "vf_coef": 0.5,
+    "ent_coef": 0.0,
+    "max_grad_norm": 0.5,
+    "separate_actor_critic_grad_clipping": true,
+    "target_kl": 0.03,
+    "normalize_reward": true,
+    "normalize_observation": false,
+    "observation_clip": 10.0,
+    "log_interval": 1,
+    "seed": 2020,
+    "separate_rng_streams": true,
+    "initialization_seed": 2020,
+    "environment_seed": 5202000,
+    "policy_sampling_seed": 102020,
+    "minibatch_seed": 302020,
+    "initial_hidden_mode": "zero",
+    "rollout_heartbeat": 4,
+    "best_checkpoint_mode": "return",
+    "best_checkpoint_window": 100,
+    "best_checkpoint_min_episodes": 10,
+    "best_checkpoint_velocity_key": "mean_x_velocity",
+    "best_checkpoint_min_forward_velocity": 0.0,
+    "best_checkpoint_max_high_z_termination_rate": 1.0,
+    "best_checkpoint_max_inverted_completion_rate": 1.0,
+    "best_checkpoint_max_functional_failure_rate": 1.0
+  },
+  "actor_parameter_count": 404593,
+  "critic_parameter_count": 242433,
+  "critic_sha256": "9d7ce49cb0c96ed0ec9af1e5b73c9f93c7ee35a74182a249c5fdcc14375058ca",
+  "graph_sha256": "883b6a156e574babecdc4bc179651247d0faa6d3fee1f9d8236b6c099e94f6b2",
+  "graph_buffers_sha256": "e85e353744dba32fe3458d16832d657279d5c9e284a680b6e7f35a6cba84e952",
+  "action_storage": "pre-tanh Gaussian coordinates",
+  "observation_dim": 688,
+  "action_dim": 6,
+  "teacher_used": false,
+  "pretrained_actor_loaded": false,
+  "readout": "normalized",
+  "body": "NeuromechFly",
+  "exploration_sampling": "antithetic_environment_halves",
+  "initial_log_std": -1.0,
+  "leg_residual_l2_limit_rad": 0.25980762113533157,
+  "motor_interface": "normalized_multijoint_leg_synergy"
+}