serialexperimentsleon commited on
Commit
32832f5
·
verified ·
1 Parent(s): 0d0a9e9

Add files using upload-large-folder tool

Browse files
Files changed (6) hide show
  1. config.yaml +693 -0
  2. snapshot_100.pt +3 -0
  3. snapshot_25.pt +3 -0
  4. snapshot_50.pt +3 -0
  5. snapshot_75.pt +3 -0
  6. snapshot_last.pt +3 -0
config.yaml ADDED
@@ -0,0 +1,693 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ root_dir: /mnt/crucialSSD/datasetsSSD/fish_datasets/simulated/teleop
2
+ replay_buffer_size: 150000
3
+ replay_buffer_num_workers: 2
4
+ nstep: 3
5
+ batch_size: 128
6
+ seed: 0
7
+ dataset_shuffle_seed: ${seed}
8
+ valid_dataset_shuffle_seed: 0
9
+ train_dataset_shuffle_seed: 0
10
+ device: cuda
11
+ save_video: true
12
+ save_train_video: true
13
+ use_tb: true
14
+ use_wandb: true
15
+ wandb_run_id: 9O0bnAwmob0oQ
16
+ wandb_notes: _
17
+ eval: false
18
+ eval_on_sim: false
19
+ eval_on_sim_past_checkpoints: false
20
+ simulation: true
21
+ true_action_history: false
22
+ train_pad_after: 8
23
+ process_contact_features: ${eval}
24
+ obs_type: pixels
25
+ use_color: true
26
+ use_depth: true
27
+ use_masks: false
28
+ mask_list:
29
+ - EE_obj_mask
30
+ mask_representation: channels
31
+ segmentation_model_name: gt_segmentation
32
+ crop_hw:
33
+ - 144
34
+ - 144
35
+ crop_down_offset: 48
36
+ color_crop_type: null
37
+ depth_crop_type: null
38
+ segmask_crop_type: null
39
+ add_crop_binary_mask: false
40
+ add_coord_conv_map: false
41
+ use_context_color: false
42
+ use_context_depth: false
43
+ use_context_segmask: false
44
+ context_color_crop_type: null
45
+ context_depth_crop_type: null
46
+ context_segmask_crop_type: null
47
+ context_add_crop_binary_mask: false
48
+ context_add_coord_conv_map: false
49
+ use_contact_map: false
50
+ use_sdf_maps: false
51
+ use_normals_maps: false
52
+ which_objects: both
53
+ max_contact_prob: 1.0
54
+ max_depth: 2.0
55
+ grasped_dtc_max_value: 0.2
56
+ env_dtc_max_value: 0.4
57
+ grasped_normals_mask_max_dtc_value: 0.2
58
+ env_normals_mask_max_dtc_value: 0.4
59
+ clamp_dtc: true
60
+ dtc_adaptive_normalization: false
61
+ mask_normals_within_sdf: true
62
+ adaptive_normals_mask: true
63
+ learnable_contact_preprocess_params: false
64
+ contact_model_name: gt_contact
65
+ contact_estimation_model_ckpt_path: ~/fish_leon/contact_estimation/artifacts/175604_2/checkpoints/epoch=09-val_loss=0.00.ckpt
66
+ encoder_type: small
67
+ debug_timestamps: false
68
+ open_loop: false
69
+ action_trajectories: true
70
+ stop_after_action: false
71
+ interpolation_frequency: 25
72
+ policy_frequency: 5
73
+ wait_for_new_camera_frames: true
74
+ baseline: false
75
+ use_num_train_epochs_to_determine_lr_schedule: true
76
+ use_train_steps_over_epochs: true
77
+ grasped_book_config:
78
+ _target_: mani_skill.envs.tasks.tabletop.book_insertion.GraspedBookConfig
79
+ randomize_color: true
80
+ randomize_density: false
81
+ density_randomization_bounds:
82
+ - 650
83
+ - 850
84
+ randomize_height: true
85
+ height_randomization_bounds:
86
+ - 0.165
87
+ - 0.25
88
+ randomize_width: true
89
+ width_randomization_bounds:
90
+ - 0.03
91
+ - 0.065
92
+ randomize_length: false
93
+ length_randomization_bounds:
94
+ - 0.1
95
+ - 0.15
96
+ env_books_config:
97
+ _target_: mani_skill.envs.tasks.tabletop.book_insertion.EnvBooksConfig
98
+ num_env_books: 8
99
+ randomize_color: false
100
+ randomize_density: false
101
+ density_randomization_bounds:
102
+ - 655
103
+ - 1015
104
+ randomize_height: false
105
+ height_randomization_bounds:
106
+ - 0.2475
107
+ - 0.2525
108
+ randomize_width: false
109
+ width_randomization_bounds:
110
+ - 0.015
111
+ - 0.05
112
+ randomize_length: false
113
+ length_randomization_bounds:
114
+ - 0.15
115
+ - 0.2
116
+ shuffle_mode: none
117
+ slot_config:
118
+ _target_: mani_skill.envs.tasks.tabletop.book_insertion.SlotConfig
119
+ negative_tolerance: 0.0035
120
+ y_randomization_bounds: 0.0
121
+ left_of_book_index: 4
122
+ book_ends_config:
123
+ _target_: mani_skill.envs.tasks.tabletop.book_insertion.BookEndsConfig
124
+ mode: spring
125
+ length: 0.2
126
+ width: 0.025
127
+ height: 0.25
128
+ mass: 1.0
129
+ color: '#808080'
130
+ friction: 0.0
131
+ wall_height: 0.25
132
+ wall_length: 0.2
133
+ wall_width: 0.025
134
+ travel_limit: 0.125
135
+ joint_stiffness: 100.0
136
+ joint_damping: 20.0
137
+ record_sim_videos: true
138
+ resume_wandb_run: true
139
+ train_demo_idxs_list_or_num: -1
140
+ log_train_every_steps: 25
141
+ name_of_expert_demo: 206_sim_demos_leftof4thbook_springbookends_nograspedrand_noenvrand_slotrand_20hz_act
142
+ expert_dataset_dirpath: ${root_dir}/FISH/expert_demos/${suite.name}/${task_name}/${name_of_expert_demo}
143
+ store_dataset_in_memory: false
144
+ expert_dataset: ${expert_dataset_dirpath}/demos.zarr
145
+ action_key: action
146
+ semantic_demo_grouping_name: semantic_demo_grouping.yaml
147
+ semantic_demo_grouping: ${expert_dataset_dirpath}/${semantic_demo_grouping_name}
148
+ include_groups_list: all
149
+ expert_dataset_config: ${expert_dataset_dirpath}/demo_config.yaml
150
+ name_of_valid_demo: 206_sim_demos_leftof4thbook_springbookends_nograspedrand_noenvrand_slotrand_20hz_act
151
+ train_num_groups: 15
152
+ valid_dataset_dir: ${root_dir}/FISH/expert_demos/${suite.name}/${task_name}/${name_of_valid_demo}/demos.zarr
153
+ valid_demo_idxs_list_or_num: null
154
+ val_num_groups: 15
155
+ universal_unseen_env_seed_start: 1000000
156
+ load_bc: ${agent.load_checkpoint}
157
+ checkpoint_epoch_list:
158
+ - 99
159
+ - 199
160
+ - 299
161
+ - 399
162
+ - 499
163
+ - 599
164
+ - 699
165
+ - 799
166
+ - 899
167
+ - 999
168
+ - 1249
169
+ - 1499
170
+ - 1749
171
+ - 1999
172
+ - 2999
173
+ - 3999
174
+ - 4999
175
+ - 5999
176
+ - 6999
177
+ - 7999
178
+ - 8999
179
+ - 9999
180
+ snapshot_root_dir: /mnt/bighdd/fish_contact_backup
181
+ save_snapshot: true
182
+ save_last_snapshot: true
183
+ save_snapshot_when_done: true
184
+ top_k_checkpoints: 10
185
+ save_snapshot_link_to_weights_dir: deprecated
186
+ restore_missing_checkpoint: null
187
+ restore_and_eval_all_missing_checkpoints: false
188
+ bc_regularize: false
189
+ bc_weight_type: qfilter
190
+ username: leonmkim
191
+ hostname: grasp-login1
192
+ local_snapshot_root_dir: /mnt/bighdd/fish_contact_backup
193
+ local_snapshot_dir: ${local_snapshot_root_dir}/exp_local/${suite.name}_${obs_type}/${task_name}
194
+ experiment_dir: ./exp_local/${suite.name}_${obs_type}/${task_name}/${wandb_run_id}
195
+ agent:
196
+ _target_: agent.diffusion_policy.DiffusionPolicyAgent
197
+ name: diffusion_policy
198
+ load_checkpoint: ${eval}
199
+ device: ${device}
200
+ n_obs_steps: ${.config.policy_cfg.n_obs_steps}
201
+ suite_name: ${suite.name}
202
+ obs_type: ${obs_type}
203
+ enable_arm: ${eval}
204
+ enable_camera: ${eval}
205
+ use_tb: ${use_tb}
206
+ desired_image_shape:
207
+ - 13
208
+ - 180
209
+ - 240
210
+ orig_cam_shape:
211
+ - 3
212
+ - 240
213
+ - 320
214
+ config:
215
+ _target_: agent.diffusion_policy.DiffusionPolicyAgentConfig
216
+ compile: false
217
+ device: ${device}
218
+ cam_resize_shape: ${agent.desired_image_shape}
219
+ orig_cam_shape: ${agent.orig_cam_shape}
220
+ policy_frequency: ${policy_frequency}
221
+ interpolation_frequency: ${interpolation_frequency}
222
+ deterministic_actions: true
223
+ deterministic_actions_seed: 0
224
+ policy_cfg:
225
+ _target_: lerobot.common.policies.diffusion.configuration_diffusion.DiffusionConfig
226
+ action_config:
227
+ _target_: lerobot.common.policies.diffusion.configuration_diffusion.ActionConfig
228
+ horizon_length: ${agent.config.policy_cfg.horizon}
229
+ action_frame_expression: delta
230
+ rotation_representation: axis_angle
231
+ maniskill_prenormalized: true
232
+ output_maniskill_normalized: ${agent.config.policy_cfg.action_config.maniskill_prenormalized}
233
+ pos_upper: 0.1
234
+ pos_lower: -0.1
235
+ rot_lower: -0.1
236
+ rot_upper: 0.1
237
+ action_history_config:
238
+ _target_: lerobot.common.policies.diffusion.configuration_diffusion.ActionHistoryConfig
239
+ enable: true
240
+ history_length: ${agent.config.policy_cfg.action_history_encoder_config.history_length}
241
+ action_frame_expression: delta
242
+ action_frame: current_end_effector
243
+ rotation_representation: axis_angle
244
+ maniskill_prenormalized: true
245
+ pos_upper: 0.1
246
+ pos_lower: -0.1
247
+ rot_lower: -0.1
248
+ rot_upper: 0.1
249
+ n_obs_steps: 1
250
+ horizon: 28
251
+ n_action_steps: ${agent.config.policy_cfg.horizon}
252
+ output_shapes:
253
+ action:
254
+ - 7
255
+ input_normalization_modes:
256
+ observation.image: mean_std
257
+ observation.state: min_max
258
+ observation.action_history: min_max
259
+ output_normalization_modes:
260
+ action: min_max
261
+ vision_backbone: resnet18
262
+ crop_distractors_tlhw: null
263
+ pretrained_backbone_weights: null
264
+ transforms:
265
+ - _target_: torchvision.transforms.v2.Grayscale
266
+ num_output_channels: 3
267
+ use_group_norm: true
268
+ spatial_softmax_num_keypoints: 32
269
+ action_history_encoder_config:
270
+ _target_: lerobot.common.policies.diffusion.configuration_diffusion.Unet1dEncoderConfig
271
+ in_channels: 7
272
+ out_channels: 32
273
+ history_length: 4
274
+ kernel_size: ${agent.config.policy_cfg.kernel_size}
275
+ downsample_kernel_size: 3
276
+ downsample_stride: 2
277
+ downsample_padding: 1
278
+ down_dims:
279
+ - 256
280
+ - 512
281
+ - 1024
282
+ kernel_size: 5
283
+ n_groups: 8
284
+ diffusion_step_embed_dim: 128
285
+ use_film_scale_modulation: true
286
+ noise_scheduler_type: DDIM
287
+ beta_schedule: squaredcos_cap_v2
288
+ beta_start: 0.0001
289
+ beta_end: 0.02
290
+ prediction_type: epsilon
291
+ clip_sample: true
292
+ clip_sample_range: 1.0
293
+ num_train_timesteps: 50
294
+ num_inference_steps: 10
295
+ do_mask_loss_for_padding: false
296
+ start_horizon_at_current_step: true
297
+ input_shapes:
298
+ observation.image:
299
+ - 13
300
+ - 180
301
+ - 240
302
+ context_observation.image:
303
+ - 13
304
+ - 180
305
+ - 240
306
+ observation.state:
307
+ - 8
308
+ observation.action_history:
309
+ - 7
310
+ train_cfg:
311
+ _target_: utils.TrainConfig
312
+ lr: 0.0001
313
+ lr_scheduler: cosine
314
+ lr_warmup_steps: 500
315
+ adam_betas:
316
+ - 0.95
317
+ - 0.999
318
+ adam_eps: 1.0e-08
319
+ adam_weight_decay: 1.0e-06
320
+ grad_clip_norm: 10
321
+ offline_steps: 30000
322
+ use_amp: true
323
+ optimizer: adamw
324
+ observation_cfg:
325
+ _target_: agent.encoder.VisualFeatureSet
326
+ use_depth: ${use_depth}
327
+ use_color: ${use_color}
328
+ mask_input_dict:
329
+ _target_: agent.encoder.MaskInputDict
330
+ enable: ${use_masks}
331
+ representation: ${mask_representation}
332
+ mask_list: ${mask_list}
333
+ segmentation_model_name: ${segmentation_model_name}
334
+ crop_input_config:
335
+ _target_: agent.encoder.CropInputConfig
336
+ color_crop_type: ${color_crop_type}
337
+ depth_crop_type: ${depth_crop_type}
338
+ segmask_crop_type: ${segmask_crop_type}
339
+ crop_hw: ${crop_hw}
340
+ crop_down_offset: ${crop_down_offset}
341
+ add_crop_binary_mask: ${add_crop_binary_mask}
342
+ add_coord_conv_map: ${add_coord_conv_map}
343
+ context_input_config:
344
+ _target_: agent.encoder.ContextInputConfig
345
+ use_color: ${use_context_color}
346
+ use_depth: ${use_context_depth}
347
+ mask_input_dict:
348
+ _target_: agent.encoder.MaskInputDict
349
+ enable: ${use_context_segmask}
350
+ representation: ${mask_representation}
351
+ mask_list: ${mask_list}
352
+ segmentation_model_name: ${segmentation_model_name}
353
+ crop_input_config:
354
+ _target_: agent.encoder.CropInputConfig
355
+ color_crop_type: ${context_color_crop_type}
356
+ depth_crop_type: ${context_depth_crop_type}
357
+ segmask_crop_type: ${context_segmask_crop_type}
358
+ crop_hw: ${crop_hw}
359
+ crop_down_offset: ${crop_down_offset}
360
+ add_crop_binary_mask: ${context_add_crop_binary_mask}
361
+ add_coord_conv_map: ${context_add_coord_conv_map}
362
+ mask_soft_approx_scheduler_config:
363
+ _target_: agent.encoder.MaskSoftApproxSchedulerConfig
364
+ num_steps: 40000
365
+ initial_value: 10.0
366
+ final_value: 1000.0
367
+ interpolation_scheme: cosine
368
+ use_contact_map: ${use_contact_map}
369
+ use_sdf_maps: ${use_sdf_maps}
370
+ use_normals_maps: ${use_normals_maps}
371
+ which_objects: ${which_objects}
372
+ grasped_dtc_max_value: ${grasped_dtc_max_value}
373
+ env_dtc_max_value: ${env_dtc_max_value}
374
+ grasped_normals_mask_max_dtc_value: ${grasped_normals_mask_max_dtc_value}
375
+ env_normals_mask_max_dtc_value: ${env_normals_mask_max_dtc_value}
376
+ clamp_dtc: ${clamp_dtc}
377
+ max_contact_prob: ${max_contact_prob}
378
+ mask_normals_within_sdf: ${mask_normals_within_sdf}
379
+ dtc_adaptive_normalization: ${dtc_adaptive_normalization}
380
+ adaptive_normals_mask: ${adaptive_normals_mask}
381
+ max_depth: ${max_depth}
382
+ image_shape: ${agent.desired_image_shape}
383
+ learnable_contact_preprocess_params: ${learnable_contact_preprocess_params}
384
+ learning_rate: 0.0001
385
+ weight_decay: 0.0
386
+ contact_model_name: ${contact_model_name}
387
+ zero_centered: true
388
+ z_score_normalize_rgb: true
389
+ crop_distractors_tlhw: ${agent.config.policy_cfg.crop_distractors_tlhw}
390
+ suite:
391
+ suite: frankagym
392
+ name: frankagym
393
+ frame_stack: ${agent.n_obs_steps}
394
+ action_repeat: 1
395
+ discount: 0.99
396
+ hidden_dim: 1024
397
+ num_train_frames: 1000000
398
+ num_seed_frames: 0
399
+ num_train_epochs: 5000
400
+ num_train_steps: 30000
401
+ validate_every_epochs: 100
402
+ validate_every_steps: 25
403
+ eval_on_sim_every_epochs: 100
404
+ eval_on_sim_every_steps: 50
405
+ validate_diffusion_on_action_loss_every_epochs: 500
406
+ validate_diffusion_on_action_loss_every_steps: 25
407
+ eval_on_sim_every_epochs_diffusion: 500
408
+ eval_on_sim_every_steps_diffusion: 1000
409
+ train_eval_diffusion_on_action_loss_every_epochs: 500
410
+ train_eval_diffusion_on_action_loss_every_steps: 25
411
+ check_topk_every_epochs: 10
412
+ check_topk_every_steps: 25
413
+ save_snapshot_every_epochs: 5000
414
+ save_snapshot_every_steps: 200
415
+ num_seen_eval_envs: 0
416
+ num_unseen_eval_envs: 1
417
+ evals_per_env: 1
418
+ eval_every_frames: 2000
419
+ num_eval_episodes: 5
420
+ save_snapshot: true
421
+ wait_for_user_to_start_episode: true
422
+ task_make_fn:
423
+ _target_: suite.frankagym.make
424
+ name: ${task_name}
425
+ height: 240
426
+ width: 320
427
+ frame_stack: ${suite.frame_stack}
428
+ action_repeat: ${suite.action_repeat}
429
+ seed: ${seed}
430
+ enable_arm: ${agent.enable_arm}
431
+ enable_gripper: ${enable_gripper}
432
+ start_with_gripper_open: ${start_with_gripper_open}
433
+ enable_camera: ${agent.enable_camera}
434
+ path_to_depth_extrinsics: ${path_to_depth_extrinsics}
435
+ contact_estimation_model_ckpt_path: ${contact_estimation_model_ckpt_path}
436
+ x_limit: ${x_limit}
437
+ y_limit: ${y_limit}
438
+ z_limit: ${z_limit}
439
+ device: ${device}
440
+ interpolation_frequency: ${interpolation_frequency}
441
+ policy_frequency: ${policy_frequency}
442
+ debug_timestamps: ${debug_timestamps}
443
+ stop_after_action: ${stop_after_action}
444
+ open_loop: ${open_loop}
445
+ wait_for_new_camera_frames: ${wait_for_new_camera_frames}
446
+ action_key: ${action_key}
447
+ action_trajectory_horizon: ${agent.config.policy_cfg.horizon}
448
+ action_trajectories: ${action_trajectories}
449
+ path_to_zarr_dataset: ${expert_dataset}
450
+ agent_policy_cfg:
451
+ _target_: agent.diffusion_policy.DiffusionPolicyAgentConfig
452
+ compile: false
453
+ device: ${device}
454
+ cam_resize_shape: ${agent.desired_image_shape}
455
+ orig_cam_shape: ${agent.orig_cam_shape}
456
+ policy_frequency: ${policy_frequency}
457
+ interpolation_frequency: ${interpolation_frequency}
458
+ deterministic_actions: true
459
+ deterministic_actions_seed: 0
460
+ policy_cfg:
461
+ _target_: lerobot.common.policies.diffusion.configuration_diffusion.DiffusionConfig
462
+ action_config:
463
+ _target_: lerobot.common.policies.diffusion.configuration_diffusion.ActionConfig
464
+ horizon_length: ${agent.config.policy_cfg.horizon}
465
+ action_frame_expression: delta
466
+ rotation_representation: axis_angle
467
+ maniskill_prenormalized: true
468
+ output_maniskill_normalized: ${agent.config.policy_cfg.action_config.maniskill_prenormalized}
469
+ pos_upper: 0.1
470
+ pos_lower: -0.1
471
+ rot_lower: -0.1
472
+ rot_upper: 0.1
473
+ action_history_config:
474
+ _target_: lerobot.common.policies.diffusion.configuration_diffusion.ActionHistoryConfig
475
+ enable: false
476
+ history_length: ${agent.config.policy_cfg.action_history_encoder_config.history_length}
477
+ action_frame_expression: delta
478
+ action_frame: current_end_effector
479
+ rotation_representation: axis_angle
480
+ maniskill_prenormalized: true
481
+ pos_upper: 0.1
482
+ pos_lower: -0.1
483
+ rot_lower: -0.1
484
+ rot_upper: 0.1
485
+ n_obs_steps: 1
486
+ horizon: 28
487
+ n_action_steps: ${agent.config.policy_cfg.horizon}
488
+ output_shapes:
489
+ action:
490
+ - 7
491
+ input_normalization_modes:
492
+ observation.image: mean_std
493
+ observation.state: min_max
494
+ observation.action_history: min_max
495
+ output_normalization_modes:
496
+ action: min_max
497
+ vision_backbone: resnet18
498
+ crop_distractors_tlhw: null
499
+ pretrained_backbone_weights: null
500
+ transforms:
501
+ - _target_: torchvision.transforms.v2.Grayscale
502
+ num_output_channels: 3
503
+ use_group_norm: true
504
+ spatial_softmax_num_keypoints: 32
505
+ action_history_encoder_config:
506
+ _target_: lerobot.common.policies.diffusion.configuration_diffusion.Unet1dEncoderConfig
507
+ in_channels: 7
508
+ out_channels: 32
509
+ history_length: 4
510
+ kernel_size: ${agent.config.policy_cfg.kernel_size}
511
+ downsample_kernel_size: 3
512
+ downsample_stride: 2
513
+ downsample_padding: 1
514
+ down_dims:
515
+ - 256
516
+ - 512
517
+ - 1024
518
+ kernel_size: 5
519
+ n_groups: 8
520
+ diffusion_step_embed_dim: 128
521
+ use_film_scale_modulation: true
522
+ noise_scheduler_type: DDIM
523
+ beta_schedule: squaredcos_cap_v2
524
+ beta_start: 0.0001
525
+ beta_end: 0.02
526
+ prediction_type: epsilon
527
+ clip_sample: true
528
+ clip_sample_range: 1.0
529
+ num_train_timesteps: 50
530
+ num_inference_steps: 10
531
+ do_mask_loss_for_padding: false
532
+ start_horizon_at_current_step: true
533
+ input_shapes:
534
+ observation.image:
535
+ - 13
536
+ - 180
537
+ - 240
538
+ context_observation.image:
539
+ - 13
540
+ - 180
541
+ - 240
542
+ observation.state:
543
+ - 8
544
+ observation.action_history:
545
+ - 7
546
+ train_cfg:
547
+ _target_: utils.TrainConfig
548
+ lr: 0.0001
549
+ lr_scheduler: cosine
550
+ lr_warmup_steps: 500
551
+ adam_betas:
552
+ - 0.95
553
+ - 0.999
554
+ adam_eps: 1.0e-08
555
+ adam_weight_decay: 1.0e-06
556
+ grad_clip_norm: 10
557
+ offline_steps: ${num_train_frames_diffusion}
558
+ use_amp: true
559
+ optimizer: adamw
560
+ observation_cfg:
561
+ _target_: agent.encoder.VisualFeatureSet
562
+ use_depth: ${use_depth}
563
+ use_color: ${use_color}
564
+ mask_input_dict:
565
+ _target_: agent.encoder.MaskInputDict
566
+ enable: ${use_masks}
567
+ representation: ${mask_representation}
568
+ mask_list: ${mask_list}
569
+ segmentation_model_name: ${segmentation_model_name}
570
+ crop_input_config:
571
+ _target_: agent.encoder.CropInputConfig
572
+ color_crop_type: ${color_crop_type}
573
+ depth_crop_type: ${depth_crop_type}
574
+ segmask_crop_type: ${segmask_crop_type}
575
+ crop_hw: ${crop_hw}
576
+ crop_down_offset: ${crop_down_offset}
577
+ add_crop_binary_mask: ${add_crop_binary_mask}
578
+ add_coord_conv_map: ${add_coord_conv_map}
579
+ context_input_config:
580
+ _target_: agent.encoder.ContextInputConfig
581
+ use_color: ${use_context_color}
582
+ use_depth: ${use_context_depth}
583
+ mask_input_dict:
584
+ _target_: agent.encoder.MaskInputDict
585
+ enable: ${use_context_segmask}
586
+ representation: ${mask_representation}
587
+ mask_list: ${mask_list}
588
+ segmentation_model_name: ${segmentation_model_name}
589
+ crop_input_config:
590
+ _target_: agent.encoder.CropInputConfig
591
+ color_crop_type: ${context_color_crop_type}
592
+ depth_crop_type: ${context_depth_crop_type}
593
+ segmask_crop_type: ${context_segmask_crop_type}
594
+ crop_hw: ${crop_hw}
595
+ crop_down_offset: ${crop_down_offset}
596
+ add_crop_binary_mask: ${context_add_crop_binary_mask}
597
+ add_coord_conv_map: ${context_add_coord_conv_map}
598
+ mask_soft_approx_scheduler_config:
599
+ _target_: agent.encoder.MaskSoftApproxSchedulerConfig
600
+ num_steps: 40000
601
+ initial_value: 10.0
602
+ final_value: 1000.0
603
+ interpolation_scheme: cosine
604
+ use_contact_map: ${use_contact_map}
605
+ use_sdf_maps: ${use_sdf_maps}
606
+ use_normals_maps: ${use_normals_maps}
607
+ which_objects: ${which_objects}
608
+ grasped_dtc_max_value: ${grasped_dtc_max_value}
609
+ env_dtc_max_value: ${env_dtc_max_value}
610
+ grasped_normals_mask_max_dtc_value: ${grasped_normals_mask_max_dtc_value}
611
+ env_normals_mask_max_dtc_value: ${env_normals_mask_max_dtc_value}
612
+ clamp_dtc: ${clamp_dtc}
613
+ max_contact_prob: ${max_contact_prob}
614
+ mask_normals_within_sdf: ${mask_normals_within_sdf}
615
+ dtc_adaptive_normalization: ${dtc_adaptive_normalization}
616
+ adaptive_normals_mask: ${adaptive_normals_mask}
617
+ max_depth: ${max_depth}
618
+ image_shape: ${agent.desired_image_shape}
619
+ learnable_contact_preprocess_params: ${learnable_contact_preprocess_params}
620
+ learning_rate: 0.0001
621
+ weight_decay: 0.0
622
+ contact_model_name: ${contact_model_name}
623
+ zero_centered: true
624
+ z_score_normalize_rgb: true
625
+ crop_distractors_tlhw: ${agent.config.policy_cfg.crop_distractors_tlhw}
626
+ true_action_history: ${true_action_history}
627
+ num_train_frames_bc: 50000
628
+ num_train_frames_drq: 1100000
629
+ stddev_schedule_drq: linear(1.0,0.1,100000)
630
+ task_name: FrankaInsertion-v1
631
+ num_train_frames_vinn: 25000
632
+ num_train_frames_diffusion: 1000000
633
+ num_train_epochs_bc: 5000
634
+ num_train_epochs_diffusion: 300
635
+ num_train_steps_diffusion: 30000
636
+ validate_every_epochs_bc: 5
637
+ validate_every_epochs_diffusion: 1
638
+ validate_every_steps_diffusion: 25
639
+ validate_diffusion_on_action_loss_every_epochs: 1
640
+ validate_diffusion_on_action_loss_every_steps: 25
641
+ train_eval_diffusion_on_action_loss_every_epochs: 1
642
+ train_eval_diffusion_on_action_loss_every_steps: 25
643
+ eval_on_sim_every_epochs_diffusion: 1
644
+ eval_on_sim_every_steps_diffusion: 50
645
+ check_topk_every_epochs: 5
646
+ check_topk_every_steps: 1000
647
+ check_topk_every_epochs_diffusion: 1
648
+ check_topk_every_steps_diffusion: 25
649
+ save_snapshot_every_epochs_diffusion: 5
650
+ save_snapshot_every_steps_diffusion: 200
651
+ x_limit:
652
+ - 0.2
653
+ - 0.7
654
+ y_limit:
655
+ - -0.4
656
+ - 0.4
657
+ z_limit:
658
+ - -0.05
659
+ - 0.55
660
+ home_displacement:
661
+ - 0.55
662
+ - 0.0
663
+ - 0.55
664
+ - 180.0
665
+ - 0.0
666
+ - 0.0
667
+ enable_gripper: true
668
+ start_with_gripper_open: true
669
+ offset_mask:
670
+ - 1
671
+ - 1
672
+ - 1
673
+ - 1
674
+ - 1
675
+ - 1
676
+ path_to_depth_extrinsics: ~/fish_leon/FISH/cfgs/camera_poses/camera_poses_L515/20240904-122305/color_tf_world.npy
677
+ test:
678
+ username: leonmkim
679
+ hostname: grasp-login1
680
+ name_of_expert_demo: 112_240x320_all_twodim_left_to_right_annotated_start_idx_5hz_zstd7_EE_pxl_coords_expert_demos_imp_act
681
+ root_dir: ~/fish_leon
682
+ expert_dataset_dirpath: ${test.root_dir}/FISH/expert_demos/${suite.name}/${task_name}/${test.name_of_expert_demo}
683
+ expert_dataset: ${test.expert_dataset_dirpath}/demos.zarr
684
+ semantic_demo_grouping_name: semantic_demo_grouping.yaml
685
+ semantic_demo_grouping: ${test.expert_dataset_dirpath}/${test.semantic_demo_grouping_name}
686
+ include_groups_list: all
687
+ batch_size: 128
688
+ store_dataset_in_memory: false
689
+ use_tb: true
690
+ local_snapshot_root_dir: /mnt/bighdd/fish_contact_backup
691
+ local_snapshot_dir: ${test.local_snapshot_root_dir}/exp_local/${suite.name}_${obs_type}/${task_name}
692
+ resume_wandb_run: false
693
+ feature_type: 180x240_1_RGB_D_2.0_acthist_hst4_out32_dwnkrnl3_dwnstrd2_dwnpd1
snapshot_100.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e948e631e8068c309cb36a1b1f72339b7aa396f2973df948e5af59a0a4e525f9
3
+ size 908717855
snapshot_25.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cc80f5fa4f073970a1a152a631a93a0f6616c445a335c9eeaab3f15ea49e7404
3
+ size 908717471
snapshot_50.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e2fc53d27b53d5d191f6db6b1e479853adf95e61f02c1530d36884a092efdb61
3
+ size 908717599
snapshot_75.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2dd47c7976fd3963f891113d123989d59d73e106d22396d8f54c62c4e58ba859
3
+ size 908717727
snapshot_last.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e948e631e8068c309cb36a1b1f72339b7aa396f2973df948e5af59a0a4e525f9
3
+ size 908717855