Upload folder using huggingface_hub

logs/robocasa__bread/train_dp-0112_demos25/seed0/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94d8c40f82b7b6fbe31fb2e559191f2ad3c384a91fff5758f7954b7949ab9a1b
+size 347078466

logs/robocasa__bread/train_dp-0112_demos25/seed0/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__bread/train_dp-0112_demos25/seed0
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 25
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 0
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__bread
+time_limit: 200
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 300000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__bread/train_dp-0112_demos25
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__bread/train_dp-0112_demos25/seed36/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e13cf1a01a47cdacb8958bb726b3a4f6aa9af7729552453cf9c8073981d120b1
+size 347078466

logs/robocasa__bread/train_dp-0112_demos25/seed36/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__bread/train_dp-0112_demos25/seed36
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 25
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 36
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__bread
+time_limit: 200
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 300000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__bread/train_dp-0112_demos25
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__bread/train_dp-0112_demos25/seed72/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc7613410a242b2f7016f171bf3cb3c9bb48349daab64a27a3e971d413311df0
+size 347078466

logs/robocasa__bread/train_dp-0112_demos25/seed72/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__bread/train_dp-0112_demos25/seed72
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 25
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 72
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__bread
+time_limit: 200
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 300000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__bread/train_dp-0112_demos25
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__breadcanrandom/train_dp-0112_demos100/seed0/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b507329eee26eb9a66b44dd38fac267e7c1fb657471cb60295c026752c04766b
+size 347078466

logs/robocasa__breadcanrandom/train_dp-0112_demos100/seed0/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__breadcanrandom/train_dp-0112_demos100/seed0
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 100
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 0
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__breadcanrandom
+time_limit: 400
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 500000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__breadcanrandom/train_dp-0112_demos100
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__breadcanrandom/train_dp-0112_demos100/seed36/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5287823b37e63ed60ccfef714768f7bf036f996f1b9551db7c2bb8dbd4d52e2
+size 347078466

logs/robocasa__breadcanrandom/train_dp-0112_demos100/seed36/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__breadcanrandom/train_dp-0112_demos100/seed36
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 100
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 36
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__breadcanrandom
+time_limit: 400
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 500000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__breadcanrandom/train_dp-0112_demos100
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__breadcanrandom/train_dp-0112_demos100/seed72/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99536cb90cb1681505e312c856130d895019f34fe82e422e29354f95f42f65ba
+size 347078466

logs/robocasa__breadcanrandom/train_dp-0112_demos100/seed72/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__breadcanrandom/train_dp-0112_demos100/seed72
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 100
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 72
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__breadcanrandom
+time_limit: 400
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 500000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__breadcanrandom/train_dp-0112_demos100
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__cereal/train_dp-0112_demos100/seed0/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__cereal/train_dp-0112_demos100/seed0
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 100
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 0
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__cereal
+time_limit: 150
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 500000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__cereal/train_dp-0112_demos100
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__cereal/train_dp-0112_demos100/seed36/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__cereal/train_dp-0112_demos100/seed36
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 100
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 36
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__cereal
+time_limit: 150
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 500000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__cereal/train_dp-0112_demos100
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__cereal/train_dp-0112_demos100/seed72/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__cereal/train_dp-0112_demos100/seed72
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 100
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 72
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__cereal
+time_limit: 150
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 500000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__cereal/train_dp-0112_demos100
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__door/train_dp-0112_demos10/seed0/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba10f8931b921760e26bdaa03148f888d694187932addcb37b9096e7856de832
+size 347078466

logs/robocasa__door/train_dp-0112_demos10/seed0/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__door/train_dp-0112_demos10/seed0
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 10
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 0
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__door
+time_limit: 200
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 100000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__door/train_dp-0112_demos10
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__door/train_dp-0112_demos10/seed36/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:183b9f4d112235e6380350e67dffead8325fe0d3799c5a6ab3eaf87315501fb8
+size 347078466

logs/robocasa__door/train_dp-0112_demos10/seed36/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__door/train_dp-0112_demos10/seed36
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 10
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 36
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__door
+time_limit: 200
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 100000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__door/train_dp-0112_demos10
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__door/train_dp-0112_demos10/seed72/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:234bda4906e51447ce4a4625ae3ff9b4b6881aa7e686122b21f9ddfb77a8a6d3
+size 347078466

logs/robocasa__door/train_dp-0112_demos10/seed72/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__door/train_dp-0112_demos10/seed72
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 10
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 72
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__door
+time_limit: 200
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 100000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__door/train_dp-0112_demos10
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__door/train_dp-0112_demos5/seed0/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91e397b9ba4bfd01738e552de11f7475c42fb2631858c3bb604dda7c2f6016a0
+size 347078466

logs/robocasa__door/train_dp-0112_demos5/seed0/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__door/train_dp-0112_demos5/seed0
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 5
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 0
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__door
+time_limit: 200
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 100000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__door/train_dp-0112_demos5
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__door/train_dp-0112_demos5/seed36/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87bad9f7c2f445e6d19be1ccc7e67006090d6c9ab8286cfc0e373d01f6ab06a1
+size 347078466

logs/robocasa__door/train_dp-0112_demos5/seed36/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__door/train_dp-0112_demos5/seed36
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 5
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 36
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__door
+time_limit: 200
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 100000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__door/train_dp-0112_demos5
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__door/train_dp-0112_demos5/seed72/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53aa371644c1326a577f984f8a16aa8fda90287033e96058cc4a486398858141
+size 347078466

logs/robocasa__door/train_dp-0112_demos5/seed72/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__door/train_dp-0112_demos5/seed72
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 5
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 72
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__door
+time_limit: 200
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 100000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__door/train_dp-0112_demos5
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__nutassemblyround/train_dp-0112_demos100/seed0/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__nutassemblyround/train_dp-0112_demos100/seed0
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 100
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 0
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__nutassemblyround
+time_limit: 300
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 500000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__nutassemblyround/train_dp-0112_demos100
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__nutassemblyround/train_dp-0112_demos100/seed36/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a415dc1a03b311796fa291231209c463746416580320d3456c028357fdee520b
+size 347078466

logs/robocasa__nutassemblyround/train_dp-0112_demos100/seed36/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__nutassemblyround/train_dp-0112_demos100/seed36
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 100
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 36
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__nutassemblyround
+time_limit: 300
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 500000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__nutassemblyround/train_dp-0112_demos100
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__nutassemblyround/train_dp-0112_demos100/seed72/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0250b398dae6047fcbf528b621b1f70a876142e366b515b0556ca9d5e52d7fe8
+size 347078466

logs/robocasa__nutassemblyround/train_dp-0112_demos100/seed72/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__nutassemblyround/train_dp-0112_demos100/seed72
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 100
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 72
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__nutassemblyround
+time_limit: 300
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 500000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__nutassemblyround/train_dp-0112_demos100
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__stack/train_dp-0112_demos25/seed0/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7037e64354f7ac9a624a85e196cc10291e0082074c99a26940dccd317afb9a9a
+size 347078466

logs/robocasa__stack/train_dp-0112_demos25/seed0/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__stack/train_dp-0112_demos25/seed0
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 25
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 0
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__stack
+time_limit: 150
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 300000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__stack/train_dp-0112_demos25
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__stack/train_dp-0112_demos25/seed36/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d90bf9148db8fbb4c993ed1f7fd25c46657e80ba4165cffce7d466a47137cbba
+size 347078466

logs/robocasa__stack/train_dp-0112_demos25/seed36/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__stack/train_dp-0112_demos25/seed36
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 25
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 36
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__stack
+time_limit: 150
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 300000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__stack/train_dp-0112_demos25
+wandb_project: Recovery
+weight_decay: 0.0

logs/robocasa__stack/train_dp-0112_demos25/seed72/DP_Pretrain_base_policy_latest.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:302f0a3edc7c824e326cf5954f55eaa6fd6afebb9a21f0fcc3c75f0a144f9d9a
+size 347078466

logs/robocasa__stack/train_dp-0112_demos25/seed72/config.yaml

@@ -0,0 +1,114 @@
+act: SiLU
+action_dim: -1
+action_horizon: 1
+action_repeat: 3
+batch_length: 32
+batch_size: 16
+compile: true
+cont_head: {layers: 2, loss_scale: 1.0, outscale: 1.0}
+critic: {dist: normal_std_fixed, eps: 1e-05, grad_clip: 100.0, layers: 2, lr: 3e-05,
+  num_models: 5, num_subsample: 2, outscale: 0.0, slow_target: true, slow_target_fraction: 0.02,
+  slow_target_update: 1}
+datadir: datasets/robocasa_datasets
+debug: false
+decoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, cnn_sigmoid: false, image_dist: mse,
+  kernel_size: 4, minres: 4, mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024,
+  norm: true, outscale: 1.0, vector_dist: symlog_mse}
+deterministic_run: false
+device: cuda:0
+discount: 0.99
+discount_lambda: 0.95
+done_mode: 1
+dp:
+  ac_chunk: 8
+  action_space: deltas
+  batch_size: 128
+  debug: false
+  device: cuda:0
+  devices: 1
+  eval_freq: 10000
+  img_chunk: 2
+  log_freq: 100
+  lr: 0.0001
+  max_iterations: 500000
+  num_cams: 2
+  pretrained_ckpt: ''
+  print_config: false
+  schedule_freq: 1
+  seed: 292285
+  shared_mlp: []
+  train_steps: 30000
+  train_transform: gpu_medium
+  use_if_ckpt_present: false
+dyn_deter: 512
+dyn_discrete: 32
+dyn_hidden: 512
+dyn_mean_act: none
+dyn_min_std: 0.1
+dyn_rec_depth: 1
+dyn_scale: 0.5
+dyn_std_act: sigmoid2
+dyn_stoch: 32
+encoder: {act: SiLU, cnn_depth: 32, cnn_keys: .*image.*, kernel_size: 4, minres: 4,
+  mlp_keys: .*state.*, mlp_layers: 5, mlp_units: 1024, norm: true, symlog_inputs: true}
+env_max_steps: {bread: 300000.0, breadcanfixed: 500000.0, breadcanrandom: 500000.0,
+  cereal: 500000.0, door: 100000.0, nutassemblyround: 500000.0, stack: 300000.0}
+env_time_limits: {bread: 200, breadcanfixed: 400, breadcanrandom: 400, cereal: 150,
+  door: 200, nutassemblyround: 300, stack: 150}
+eval_num_runs: 50
+generate_highres_eval: false
+grad_clip: 1000
+grad_heads: [decoder, cont]
+high_res_render: false
+highres_img_size: 1024
+imag_horizon: 7
+image_size: 64
+initial: learned
+kl_free: 1.0
+log_every: 50
+logdir: scratch_dir//logs/robocasa__stack/train_dp-0112_demos25/seed72
+model_lr: 0.0001
+mppi: {abs_residual: 0.2, action_l2_cost: 0, discount: 0.99, horizon: 8, init_std: 0.05,
+  iterations: 6, max_std: 0.1, min_std: 0.01, num_elites: 32, num_samples: 256, pretrained_ckpt: '',
+  temperature: 0.5, uncertainty_cost: 1.0}
+norm: True'
+normalize_state_actions: false
+num_buffer_transitions: 100000
+num_envs: 10
+num_exp_trajs: 25
+num_exp_val_trajs: 1
+obs_horizon: 2
+opt: adam
+opt_eps: 1e-08
+precision: 32
+pred_horizon: 8
+rep_scale: 0.1
+residual_actor: {absmax: null, dist: normal, entropy: 0.0003, eps: 1e-05, grad_clip: 100.0,
+  layers: 2, lr: 3e-05, max_std: 0.1, min_std: 0.01, outscale: 1.0, std: learned,
+  temp: 0.1, unimix_ratio: 0.01}
+reward_EMA: true
+reward_head: {dist: normal_std_fixed, layers: 2, outscale: 0.0}
+scratch_dir: scratch_dir/
+seed: 72
+set:
+- [image_size, '64']
+state_dim: -1
+state_only: false
+task: robocasa__stack
+time_limit: 150
+train_dp_mppi: false
+train_dp_mppi_params: {data_collect_noise_std: 0.1, discrim_state_only: true, dp_expert_buffer_ratio: 0.5,
+  eval_every_round: 10, force_warmup: true, min_env_steps_per_round: 3500, n_dp_train_itrs: 1000,
+  n_dp_traj_buffer_size: 64, n_env_steps: 300000.0, n_traj_to_relabel_per_round: 64,
+  rounds_train_ratio: 1.5, upate_discrim_every: 100, update_dp_every: 10, use_discrim: true,
+  warmstart_percentage_env_steps: 0.2, warmstart_train_ratio: 1.5}
+unimix_ratio: 0.01
+units: 512
+update_dp: false
+use_wandb: true
+visualize_eval: true
+viz_expert_buffer: false
+wandb_entity: njuai
+wandb_exp_name: robocasa__stack/train_dp-0112_demos25
+wandb_project: Recovery
+weight_decay: 0.0