MikelRubi commited on
Commit
fceb1e2
·
verified ·
1 Parent(s): 69d2800

Add pick_v3_1-2-pieces from IkerAzurmendi (act-pick-v3-1-2-pieces-6000, act-pick-v3-1-2-pieces-12000, act-pick-v3-1-2-pieces-18000, act-pick-v3-1-2-pieces-24000, act-pick-v3-1-2-pieces-30000)

Browse files
Files changed (45) hide show
  1. pick_v3_1-2-pieces/checkpoints/006000/dataset_features.json +88 -0
  2. pick_v3_1-2-pieces/checkpoints/006000/model_card_original.md +49 -0
  3. pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/config.json +71 -0
  4. pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/model.safetensors +3 -0
  5. pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/policy_postprocessor.json +32 -0
  6. pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors +3 -0
  7. pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/policy_preprocessor.json +64 -0
  8. pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors +3 -0
  9. pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/train_config.json +533 -0
  10. pick_v3_1-2-pieces/checkpoints/012000/dataset_features.json +88 -0
  11. pick_v3_1-2-pieces/checkpoints/012000/model_card_original.md +49 -0
  12. pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/config.json +71 -0
  13. pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/model.safetensors +3 -0
  14. pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/policy_postprocessor.json +32 -0
  15. pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors +3 -0
  16. pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/policy_preprocessor.json +64 -0
  17. pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors +3 -0
  18. pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/train_config.json +533 -0
  19. pick_v3_1-2-pieces/checkpoints/018000/dataset_features.json +88 -0
  20. pick_v3_1-2-pieces/checkpoints/018000/model_card_original.md +40 -0
  21. pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/config.json +71 -0
  22. pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/model.safetensors +3 -0
  23. pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/policy_postprocessor.json +32 -0
  24. pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors +3 -0
  25. pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/policy_preprocessor.json +64 -0
  26. pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors +3 -0
  27. pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/train_config.json +533 -0
  28. pick_v3_1-2-pieces/checkpoints/024000/dataset_features.json +88 -0
  29. pick_v3_1-2-pieces/checkpoints/024000/model_card_original.md +49 -0
  30. pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/config.json +71 -0
  31. pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/model.safetensors +3 -0
  32. pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/policy_postprocessor.json +32 -0
  33. pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors +3 -0
  34. pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/policy_preprocessor.json +64 -0
  35. pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors +3 -0
  36. pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/train_config.json +533 -0
  37. pick_v3_1-2-pieces/checkpoints/030000/dataset_features.json +88 -0
  38. pick_v3_1-2-pieces/checkpoints/030000/model_card_original.md +49 -0
  39. pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/config.json +71 -0
  40. pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/model.safetensors +3 -0
  41. pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/policy_postprocessor.json +32 -0
  42. pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors +3 -0
  43. pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/policy_preprocessor.json +64 -0
  44. pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors +3 -0
  45. pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/train_config.json +533 -0
pick_v3_1-2-pieces/checkpoints/006000/dataset_features.json ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action": {
3
+ "dtype": "float32",
4
+ "names": [
5
+ "shoulder_pan.pos",
6
+ "shoulder_lift.pos",
7
+ "elbow_flex.pos",
8
+ "wrist_flex.pos",
9
+ "wrist_roll.pos",
10
+ "gripper.pos"
11
+ ],
12
+ "shape": [
13
+ 6
14
+ ]
15
+ },
16
+ "observation.images.front": {
17
+ "dtype": "video",
18
+ "shape": [
19
+ 480,
20
+ 640,
21
+ 3
22
+ ],
23
+ "names": [
24
+ "height",
25
+ "width",
26
+ "channels"
27
+ ],
28
+ "info": {
29
+ "is_depth_map": false,
30
+ "video.height": 480,
31
+ "video.width": 640,
32
+ "video.codec": "h264",
33
+ "video.pix_fmt": "yuv420p",
34
+ "video.fps": 30,
35
+ "video.channels": 3,
36
+ "has_audio": false,
37
+ "video.g": 2,
38
+ "video.crf": 30,
39
+ "video.preset": "p1",
40
+ "video.fast_decode": 0,
41
+ "video.video_backend": "pyav",
42
+ "video.extra_options": {}
43
+ }
44
+ },
45
+ "observation.images.wrist": {
46
+ "dtype": "video",
47
+ "shape": [
48
+ 480,
49
+ 640,
50
+ 3
51
+ ],
52
+ "names": [
53
+ "height",
54
+ "width",
55
+ "channels"
56
+ ],
57
+ "info": {
58
+ "is_depth_map": false,
59
+ "video.height": 480,
60
+ "video.width": 640,
61
+ "video.codec": "h264",
62
+ "video.pix_fmt": "yuv420p",
63
+ "video.fps": 30,
64
+ "video.channels": 3,
65
+ "has_audio": false,
66
+ "video.g": 2,
67
+ "video.crf": 30,
68
+ "video.preset": "p1",
69
+ "video.fast_decode": 0,
70
+ "video.video_backend": "pyav",
71
+ "video.extra_options": {}
72
+ }
73
+ },
74
+ "observation.state": {
75
+ "dtype": "float32",
76
+ "names": [
77
+ "shoulder_pan.pos",
78
+ "shoulder_lift.pos",
79
+ "elbow_flex.pos",
80
+ "wrist_flex.pos",
81
+ "wrist_roll.pos",
82
+ "gripper.pos"
83
+ ],
84
+ "shape": [
85
+ 6
86
+ ]
87
+ }
88
+ }
pick_v3_1-2-pieces/checkpoints/006000/model_card_original.md ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: lerobot
4
+ pipeline_tag: robotics
5
+ tags:
6
+ - lerobot
7
+ - act
8
+ - so101
9
+ - pick
10
+ ---
11
+
12
+ # act-pick-v3-1-2-pieces-6000
13
+
14
+ The v2 recipe plus the two new evening sessions of 24 Oct (third round, v3), 30,000-step run. Checkpoint at 6,000 steps. This is not the best checkpoint of its run (the best one is at 18,000 steps); it is published for completeness.
15
+
16
+ ACT policy for the SO-101 "pick a piece from the crate and hold it up to the camera" task. Inputs: front camera, wrist camera and the six joint positions; output: a chunk of 90 future joint targets at 30 fps.
17
+
18
+ ## Training data
19
+
20
+ 331 training episodes, 30 fps only, one or two pieces in the crate at the start (every 24 Oct session, plus 12 episodes of 23 Oct), 5 to 15 s long, start gap at most 20 degrees, at most 3 s idle at either end, gripper opens at least once, known bad takes removed.
21
+
22
+ Recipe: LeRobot `train()`, batch 8, chunk 90, 30 action steps per replan, default image augmentation, seed 1000.
23
+
24
+ ## Offline check on the held-out episodes
25
+
26
+ Held out: 16 episodes, 12 from `so101-pick-cube_20261024_181939` (one piece) and 4 from `so101-pick-cube_20261024_144537` (two pieces). Mean absolute joint error over the predicted chunk (lower is better) against a "stay at the current joint positions" baseline of 20.71:
27
+
28
+ | checkpoint of this run | error | below baseline |
29
+ |---|---|---|
30
+ | **6,000 steps (this one)** | **15.86** | 23% |
31
+ | 12,000 steps | 15.89 | 23% |
32
+ | 18,000 steps | 15.35 | 26% |
33
+ | 24,000 steps | 15.73 | 24% |
34
+ | 30,000 steps | 16.38 | 21% |
35
+
36
+ All runs of this round, best checkpoint of each:
37
+
38
+ | run | training episodes | best checkpoint | error | below stay-still | repo |
39
+ |---|---|---|---|---|---|
40
+ | super (all quality-passing data) | 398 | 6k | 15.16 | 27% | act-pick-v3-6000 |
41
+ | strict (6-12 s, tight quality) | 205 | 2.4k | 15.14 | 27% | act-pick-v3-strict-2400 |
42
+ | v3 recipe on 1-2 piece episodes | 331 | 18k | 15.35 | 26% | act-pick-v3-1-2-pieces-18000 |
43
+ | v2 (earlier) | 278 | 6k | 15.36 | 26% | act-pick-v2-6000 |
44
+ | chunk 60 (own baseline 16.23) | 331 | 4.8k | 11.94 | 26% | act-pick-v3-chunk60-4800 |
45
+ | one piece only | 292 | 9.6k | 15.47 | 25% | act-pick-v3-1piece-9600 |
46
+ | recent sessions only | 100 | 12k | 16.87 | 19% | act-pick-v3-recent-12000 |
47
+ | first run | 165 | 6k | 16.87 | 19% | act-pick-* |
48
+
49
+ The best checkpoints of the main runs are within noise of each other, and every run got worse after about one pass over the data. The offline error cannot say which model grasps best; none of them has been tested on the robot.
pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/config.json ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "act",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.state": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 6
9
+ ]
10
+ },
11
+ "observation.images.front": {
12
+ "type": "VISUAL",
13
+ "shape": [
14
+ 3,
15
+ 480,
16
+ 640
17
+ ]
18
+ },
19
+ "observation.images.wrist": {
20
+ "type": "VISUAL",
21
+ "shape": [
22
+ 3,
23
+ 480,
24
+ 640
25
+ ]
26
+ }
27
+ },
28
+ "output_features": {
29
+ "action": {
30
+ "type": "ACTION",
31
+ "shape": [
32
+ 6
33
+ ]
34
+ }
35
+ },
36
+ "device": "cuda",
37
+ "use_amp": false,
38
+ "use_peft": false,
39
+ "push_to_hub": false,
40
+ "repo_id": null,
41
+ "private": null,
42
+ "tags": null,
43
+ "license": null,
44
+ "pretrained_path": null,
45
+ "chunk_size": 90,
46
+ "n_action_steps": 30,
47
+ "normalization_mapping": {
48
+ "VISUAL": "MEAN_STD",
49
+ "STATE": "MEAN_STD",
50
+ "ACTION": "MEAN_STD"
51
+ },
52
+ "vision_backbone": "resnet18",
53
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
54
+ "replace_final_stride_with_dilation": false,
55
+ "pre_norm": false,
56
+ "dim_model": 512,
57
+ "n_heads": 8,
58
+ "dim_feedforward": 3200,
59
+ "feedforward_activation": "relu",
60
+ "n_encoder_layers": 4,
61
+ "n_decoder_layers": 1,
62
+ "use_vae": true,
63
+ "latent_dim": 32,
64
+ "n_vae_encoder_layers": 4,
65
+ "temporal_ensemble_coeff": null,
66
+ "dropout": 0.1,
67
+ "kl_weight": 10.0,
68
+ "optimizer_lr": 1e-05,
69
+ "optimizer_weight_decay": 0.0001,
70
+ "optimizer_lr_backbone": 1e-05
71
+ }
pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:32a7b2342b657c307122a997a102e3631eaf4deb2b3a7081d500046fba8136e8
3
+ size 206658768
pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/policy_postprocessor.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "unnormalizer_processor",
6
+ "config": {
7
+ "eps": 1e-08,
8
+ "features": {
9
+ "action": {
10
+ "type": "ACTION",
11
+ "shape": [
12
+ 6
13
+ ]
14
+ }
15
+ },
16
+ "norm_map": {
17
+ "VISUAL": "MEAN_STD",
18
+ "STATE": "MEAN_STD",
19
+ "ACTION": "MEAN_STD"
20
+ }
21
+ },
22
+ "state_file": "policy_postprocessor_step_0_unnormalizer_processor.safetensors"
23
+ },
24
+ {
25
+ "registry_name": "device_processor",
26
+ "config": {
27
+ "device": "cpu",
28
+ "float_dtype": null
29
+ }
30
+ }
31
+ ]
32
+ }
pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:931ed929c95ef00f6e334882511164d4c9d89f16bfb31eeb212ddecf99897ed7
3
+ size 7552
pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/policy_preprocessor.json ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {}
8
+ }
9
+ },
10
+ {
11
+ "registry_name": "to_batch_processor",
12
+ "config": {}
13
+ },
14
+ {
15
+ "registry_name": "device_processor",
16
+ "config": {
17
+ "device": "cuda",
18
+ "float_dtype": null
19
+ }
20
+ },
21
+ {
22
+ "registry_name": "normalizer_processor",
23
+ "config": {
24
+ "eps": 1e-08,
25
+ "features": {
26
+ "observation.state": {
27
+ "type": "STATE",
28
+ "shape": [
29
+ 6
30
+ ]
31
+ },
32
+ "observation.images.front": {
33
+ "type": "VISUAL",
34
+ "shape": [
35
+ 3,
36
+ 480,
37
+ 640
38
+ ]
39
+ },
40
+ "observation.images.wrist": {
41
+ "type": "VISUAL",
42
+ "shape": [
43
+ 3,
44
+ 480,
45
+ 640
46
+ ]
47
+ },
48
+ "action": {
49
+ "type": "ACTION",
50
+ "shape": [
51
+ 6
52
+ ]
53
+ }
54
+ },
55
+ "norm_map": {
56
+ "VISUAL": "MEAN_STD",
57
+ "STATE": "MEAN_STD",
58
+ "ACTION": "MEAN_STD"
59
+ }
60
+ },
61
+ "state_file": "policy_preprocessor_step_3_normalizer_processor.safetensors"
62
+ }
63
+ ]
64
+ }
pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:931ed929c95ef00f6e334882511164d4c9d89f16bfb31eeb212ddecf99897ed7
3
+ size 7552
pick_v3_1-2-pieces/checkpoints/006000/pretrained_model/train_config.json ADDED
@@ -0,0 +1,533 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "local/so101_act",
4
+ "root": "C:\\Users\\Usuario\\Documents\\Workspace\\LeRobot_\\demos\\06_imitation_learning\\data\\real_so101\\act_pick_super",
5
+ "episodes": [
6
+ 64,
7
+ 65,
8
+ 70,
9
+ 74,
10
+ 78,
11
+ 79,
12
+ 84,
13
+ 89,
14
+ 90,
15
+ 93,
16
+ 94,
17
+ 100,
18
+ 101,
19
+ 102,
20
+ 103,
21
+ 104,
22
+ 105,
23
+ 106,
24
+ 107,
25
+ 109,
26
+ 110,
27
+ 111,
28
+ 112,
29
+ 113,
30
+ 114,
31
+ 115,
32
+ 116,
33
+ 117,
34
+ 118,
35
+ 119,
36
+ 120,
37
+ 121,
38
+ 122,
39
+ 124,
40
+ 125,
41
+ 126,
42
+ 127,
43
+ 128,
44
+ 129,
45
+ 130,
46
+ 131,
47
+ 132,
48
+ 133,
49
+ 134,
50
+ 135,
51
+ 136,
52
+ 137,
53
+ 138,
54
+ 139,
55
+ 140,
56
+ 141,
57
+ 142,
58
+ 143,
59
+ 144,
60
+ 145,
61
+ 146,
62
+ 147,
63
+ 148,
64
+ 149,
65
+ 150,
66
+ 151,
67
+ 152,
68
+ 153,
69
+ 154,
70
+ 155,
71
+ 156,
72
+ 157,
73
+ 158,
74
+ 159,
75
+ 160,
76
+ 161,
77
+ 162,
78
+ 163,
79
+ 164,
80
+ 165,
81
+ 166,
82
+ 167,
83
+ 168,
84
+ 169,
85
+ 170,
86
+ 171,
87
+ 172,
88
+ 173,
89
+ 174,
90
+ 175,
91
+ 176,
92
+ 177,
93
+ 178,
94
+ 179,
95
+ 180,
96
+ 181,
97
+ 182,
98
+ 183,
99
+ 184,
100
+ 185,
101
+ 186,
102
+ 187,
103
+ 188,
104
+ 189,
105
+ 190,
106
+ 191,
107
+ 192,
108
+ 193,
109
+ 194,
110
+ 195,
111
+ 196,
112
+ 197,
113
+ 198,
114
+ 199,
115
+ 200,
116
+ 201,
117
+ 202,
118
+ 203,
119
+ 204,
120
+ 205,
121
+ 206,
122
+ 207,
123
+ 208,
124
+ 209,
125
+ 210,
126
+ 211,
127
+ 212,
128
+ 213,
129
+ 214,
130
+ 215,
131
+ 216,
132
+ 217,
133
+ 218,
134
+ 219,
135
+ 220,
136
+ 221,
137
+ 222,
138
+ 223,
139
+ 224,
140
+ 225,
141
+ 226,
142
+ 227,
143
+ 228,
144
+ 229,
145
+ 230,
146
+ 231,
147
+ 232,
148
+ 233,
149
+ 235,
150
+ 236,
151
+ 237,
152
+ 238,
153
+ 239,
154
+ 240,
155
+ 241,
156
+ 242,
157
+ 243,
158
+ 244,
159
+ 245,
160
+ 246,
161
+ 247,
162
+ 248,
163
+ 249,
164
+ 250,
165
+ 251,
166
+ 252,
167
+ 253,
168
+ 255,
169
+ 256,
170
+ 257,
171
+ 258,
172
+ 259,
173
+ 260,
174
+ 261,
175
+ 262,
176
+ 263,
177
+ 264,
178
+ 265,
179
+ 266,
180
+ 267,
181
+ 268,
182
+ 269,
183
+ 270,
184
+ 271,
185
+ 272,
186
+ 273,
187
+ 274,
188
+ 275,
189
+ 276,
190
+ 277,
191
+ 278,
192
+ 279,
193
+ 280,
194
+ 281,
195
+ 282,
196
+ 283,
197
+ 284,
198
+ 285,
199
+ 286,
200
+ 287,
201
+ 288,
202
+ 289,
203
+ 290,
204
+ 291,
205
+ 292,
206
+ 293,
207
+ 294,
208
+ 295,
209
+ 296,
210
+ 297,
211
+ 298,
212
+ 299,
213
+ 300,
214
+ 301,
215
+ 302,
216
+ 303,
217
+ 304,
218
+ 305,
219
+ 306,
220
+ 307,
221
+ 308,
222
+ 309,
223
+ 310,
224
+ 311,
225
+ 312,
226
+ 313,
227
+ 314,
228
+ 315,
229
+ 316,
230
+ 317,
231
+ 318,
232
+ 319,
233
+ 320,
234
+ 325,
235
+ 326,
236
+ 327,
237
+ 329,
238
+ 330,
239
+ 331,
240
+ 332,
241
+ 333,
242
+ 334,
243
+ 335,
244
+ 336,
245
+ 337,
246
+ 338,
247
+ 339,
248
+ 340,
249
+ 341,
250
+ 342,
251
+ 343,
252
+ 344,
253
+ 345,
254
+ 346,
255
+ 347,
256
+ 348,
257
+ 349,
258
+ 351,
259
+ 352,
260
+ 353,
261
+ 354,
262
+ 355,
263
+ 356,
264
+ 357,
265
+ 358,
266
+ 359,
267
+ 360,
268
+ 361,
269
+ 364,
270
+ 365,
271
+ 366,
272
+ 367,
273
+ 368,
274
+ 369,
275
+ 370,
276
+ 371,
277
+ 372,
278
+ 373,
279
+ 374,
280
+ 375,
281
+ 376,
282
+ 377,
283
+ 378,
284
+ 391,
285
+ 392,
286
+ 393,
287
+ 394,
288
+ 395,
289
+ 396,
290
+ 397,
291
+ 398,
292
+ 399,
293
+ 400,
294
+ 401,
295
+ 402,
296
+ 403,
297
+ 404,
298
+ 405,
299
+ 406,
300
+ 407,
301
+ 408,
302
+ 409,
303
+ 410,
304
+ 411,
305
+ 412,
306
+ 413,
307
+ 414,
308
+ 415,
309
+ 416,
310
+ 417,
311
+ 419,
312
+ 420,
313
+ 421,
314
+ 422,
315
+ 423,
316
+ 424,
317
+ 425,
318
+ 426,
319
+ 427,
320
+ 428,
321
+ 429,
322
+ 430,
323
+ 431,
324
+ 432,
325
+ 433,
326
+ 434,
327
+ 435,
328
+ 436,
329
+ 438,
330
+ 439,
331
+ 440,
332
+ 441,
333
+ 442,
334
+ 443,
335
+ 444,
336
+ 445
337
+ ],
338
+ "image_transforms": {
339
+ "enable": true,
340
+ "max_num_transforms": 3,
341
+ "random_order": false,
342
+ "tfs": {
343
+ "brightness": {
344
+ "weight": 1.0,
345
+ "type": "ColorJitter",
346
+ "kwargs": {
347
+ "brightness": [
348
+ 0.8,
349
+ 1.2
350
+ ]
351
+ }
352
+ },
353
+ "contrast": {
354
+ "weight": 1.0,
355
+ "type": "ColorJitter",
356
+ "kwargs": {
357
+ "contrast": [
358
+ 0.8,
359
+ 1.2
360
+ ]
361
+ }
362
+ },
363
+ "saturation": {
364
+ "weight": 1.0,
365
+ "type": "ColorJitter",
366
+ "kwargs": {
367
+ "saturation": [
368
+ 0.5,
369
+ 1.5
370
+ ]
371
+ }
372
+ },
373
+ "hue": {
374
+ "weight": 1.0,
375
+ "type": "ColorJitter",
376
+ "kwargs": {
377
+ "hue": [
378
+ -0.05,
379
+ 0.05
380
+ ]
381
+ }
382
+ },
383
+ "sharpness": {
384
+ "weight": 1.0,
385
+ "type": "SharpnessJitter",
386
+ "kwargs": {
387
+ "sharpness": [
388
+ 0.5,
389
+ 1.5
390
+ ]
391
+ }
392
+ },
393
+ "affine": {
394
+ "weight": 1.0,
395
+ "type": "RandomAffine",
396
+ "kwargs": {
397
+ "degrees": [
398
+ -5.0,
399
+ 5.0
400
+ ],
401
+ "translate": [
402
+ 0.05,
403
+ 0.05
404
+ ]
405
+ }
406
+ }
407
+ }
408
+ },
409
+ "revision": null,
410
+ "use_imagenet_stats": true,
411
+ "video_backend": "pyav",
412
+ "streaming": false
413
+ },
414
+ "env": null,
415
+ "policy": {
416
+ "type": "act",
417
+ "n_obs_steps": 1,
418
+ "input_features": {
419
+ "observation.state": {
420
+ "type": "STATE",
421
+ "shape": [
422
+ 6
423
+ ]
424
+ },
425
+ "observation.images.front": {
426
+ "type": "VISUAL",
427
+ "shape": [
428
+ 3,
429
+ 480,
430
+ 640
431
+ ]
432
+ },
433
+ "observation.images.wrist": {
434
+ "type": "VISUAL",
435
+ "shape": [
436
+ 3,
437
+ 480,
438
+ 640
439
+ ]
440
+ }
441
+ },
442
+ "output_features": {
443
+ "action": {
444
+ "type": "ACTION",
445
+ "shape": [
446
+ 6
447
+ ]
448
+ }
449
+ },
450
+ "device": "cuda",
451
+ "use_amp": false,
452
+ "use_peft": false,
453
+ "push_to_hub": false,
454
+ "repo_id": null,
455
+ "private": null,
456
+ "tags": null,
457
+ "license": null,
458
+ "pretrained_path": null,
459
+ "chunk_size": 90,
460
+ "n_action_steps": 30,
461
+ "normalization_mapping": {
462
+ "VISUAL": "MEAN_STD",
463
+ "STATE": "MEAN_STD",
464
+ "ACTION": "MEAN_STD"
465
+ },
466
+ "vision_backbone": "resnet18",
467
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
468
+ "replace_final_stride_with_dilation": false,
469
+ "pre_norm": false,
470
+ "dim_model": 512,
471
+ "n_heads": 8,
472
+ "dim_feedforward": 3200,
473
+ "feedforward_activation": "relu",
474
+ "n_encoder_layers": 4,
475
+ "n_decoder_layers": 1,
476
+ "use_vae": true,
477
+ "latent_dim": 32,
478
+ "n_vae_encoder_layers": 4,
479
+ "temporal_ensemble_coeff": null,
480
+ "dropout": 0.1,
481
+ "kl_weight": 10.0,
482
+ "optimizer_lr": 1e-05,
483
+ "optimizer_weight_decay": 0.0001,
484
+ "optimizer_lr_backbone": 1e-05
485
+ },
486
+ "output_dir": "C:\\Users\\Usuario\\Documents\\Workspace\\LeRobot_\\demos\\06_imitation_learning\\outputs\\sweep_v3",
487
+ "job_name": "act_so101_sweep_v3",
488
+ "resume": false,
489
+ "seed": 1000,
490
+ "num_workers": 4,
491
+ "batch_size": 8,
492
+ "steps": 30000,
493
+ "eval_freq": 20000,
494
+ "log_freq": 100,
495
+ "tolerance_s": 0.0001,
496
+ "save_checkpoint": true,
497
+ "save_freq": 6000,
498
+ "use_policy_training_preset": true,
499
+ "optimizer": {
500
+ "type": "adamw",
501
+ "lr": 1e-05,
502
+ "weight_decay": 0.0001,
503
+ "grad_clip_norm": 10.0,
504
+ "betas": [
505
+ 0.9,
506
+ 0.999
507
+ ],
508
+ "eps": 1e-08
509
+ },
510
+ "scheduler": null,
511
+ "eval": {
512
+ "n_episodes": 50,
513
+ "batch_size": 50,
514
+ "use_async_envs": false
515
+ },
516
+ "wandb": {
517
+ "enable": false,
518
+ "disable_artifact": false,
519
+ "project": "lerobot",
520
+ "entity": null,
521
+ "notes": null,
522
+ "run_id": null,
523
+ "mode": null
524
+ },
525
+ "peft": null,
526
+ "use_rabc": false,
527
+ "rabc_progress_path": null,
528
+ "rabc_kappa": 0.01,
529
+ "rabc_epsilon": 1e-06,
530
+ "rabc_head_mode": "sparse",
531
+ "rename_map": {},
532
+ "checkpoint_path": null
533
+ }
pick_v3_1-2-pieces/checkpoints/012000/dataset_features.json ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action": {
3
+ "dtype": "float32",
4
+ "names": [
5
+ "shoulder_pan.pos",
6
+ "shoulder_lift.pos",
7
+ "elbow_flex.pos",
8
+ "wrist_flex.pos",
9
+ "wrist_roll.pos",
10
+ "gripper.pos"
11
+ ],
12
+ "shape": [
13
+ 6
14
+ ]
15
+ },
16
+ "observation.images.front": {
17
+ "dtype": "video",
18
+ "shape": [
19
+ 480,
20
+ 640,
21
+ 3
22
+ ],
23
+ "names": [
24
+ "height",
25
+ "width",
26
+ "channels"
27
+ ],
28
+ "info": {
29
+ "is_depth_map": false,
30
+ "video.height": 480,
31
+ "video.width": 640,
32
+ "video.codec": "h264",
33
+ "video.pix_fmt": "yuv420p",
34
+ "video.fps": 30,
35
+ "video.channels": 3,
36
+ "has_audio": false,
37
+ "video.g": 2,
38
+ "video.crf": 30,
39
+ "video.preset": "p1",
40
+ "video.fast_decode": 0,
41
+ "video.video_backend": "pyav",
42
+ "video.extra_options": {}
43
+ }
44
+ },
45
+ "observation.images.wrist": {
46
+ "dtype": "video",
47
+ "shape": [
48
+ 480,
49
+ 640,
50
+ 3
51
+ ],
52
+ "names": [
53
+ "height",
54
+ "width",
55
+ "channels"
56
+ ],
57
+ "info": {
58
+ "is_depth_map": false,
59
+ "video.height": 480,
60
+ "video.width": 640,
61
+ "video.codec": "h264",
62
+ "video.pix_fmt": "yuv420p",
63
+ "video.fps": 30,
64
+ "video.channels": 3,
65
+ "has_audio": false,
66
+ "video.g": 2,
67
+ "video.crf": 30,
68
+ "video.preset": "p1",
69
+ "video.fast_decode": 0,
70
+ "video.video_backend": "pyav",
71
+ "video.extra_options": {}
72
+ }
73
+ },
74
+ "observation.state": {
75
+ "dtype": "float32",
76
+ "names": [
77
+ "shoulder_pan.pos",
78
+ "shoulder_lift.pos",
79
+ "elbow_flex.pos",
80
+ "wrist_flex.pos",
81
+ "wrist_roll.pos",
82
+ "gripper.pos"
83
+ ],
84
+ "shape": [
85
+ 6
86
+ ]
87
+ }
88
+ }
pick_v3_1-2-pieces/checkpoints/012000/model_card_original.md ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: lerobot
4
+ pipeline_tag: robotics
5
+ tags:
6
+ - lerobot
7
+ - act
8
+ - so101
9
+ - pick
10
+ ---
11
+
12
+ # act-pick-v3-1-2-pieces-12000
13
+
14
+ The v2 recipe plus the two new evening sessions of 24 Oct (third round, v3), 30,000-step run. Checkpoint at 12,000 steps. This is not the best checkpoint of its run (the best one is at 18,000 steps); it is published for completeness.
15
+
16
+ ACT policy for the SO-101 "pick a piece from the crate and hold it up to the camera" task. Inputs: front camera, wrist camera and the six joint positions; output: a chunk of 90 future joint targets at 30 fps.
17
+
18
+ ## Training data
19
+
20
+ 331 training episodes, 30 fps only, one or two pieces in the crate at the start (every 24 Oct session, plus 12 episodes of 23 Oct), 5 to 15 s long, start gap at most 20 degrees, at most 3 s idle at either end, gripper opens at least once, known bad takes removed.
21
+
22
+ Recipe: LeRobot `train()`, batch 8, chunk 90, 30 action steps per replan, default image augmentation, seed 1000.
23
+
24
+ ## Offline check on the held-out episodes
25
+
26
+ Held out: 16 episodes, 12 from `so101-pick-cube_20261024_181939` (one piece) and 4 from `so101-pick-cube_20261024_144537` (two pieces). Mean absolute joint error over the predicted chunk (lower is better) against a "stay at the current joint positions" baseline of 20.71:
27
+
28
+ | checkpoint of this run | error | below baseline |
29
+ |---|---|---|
30
+ | 6,000 steps | 15.86 | 23% |
31
+ | **12,000 steps (this one)** | **15.89** | 23% |
32
+ | 18,000 steps | 15.35 | 26% |
33
+ | 24,000 steps | 15.73 | 24% |
34
+ | 30,000 steps | 16.38 | 21% |
35
+
36
+ All runs of this round, best checkpoint of each:
37
+
38
+ | run | training episodes | best checkpoint | error | below stay-still | repo |
39
+ |---|---|---|---|---|---|
40
+ | super (all quality-passing data) | 398 | 6k | 15.16 | 27% | act-pick-v3-6000 |
41
+ | strict (6-12 s, tight quality) | 205 | 2.4k | 15.14 | 27% | act-pick-v3-strict-2400 |
42
+ | v3 recipe on 1-2 piece episodes | 331 | 18k | 15.35 | 26% | act-pick-v3-1-2-pieces-18000 |
43
+ | v2 (earlier) | 278 | 6k | 15.36 | 26% | act-pick-v2-6000 |
44
+ | chunk 60 (own baseline 16.23) | 331 | 4.8k | 11.94 | 26% | act-pick-v3-chunk60-4800 |
45
+ | one piece only | 292 | 9.6k | 15.47 | 25% | act-pick-v3-1piece-9600 |
46
+ | recent sessions only | 100 | 12k | 16.87 | 19% | act-pick-v3-recent-12000 |
47
+ | first run | 165 | 6k | 16.87 | 19% | act-pick-* |
48
+
49
+ The best checkpoints of the main runs are within noise of each other, and every run got worse after about one pass over the data. The offline error cannot say which model grasps best; none of them has been tested on the robot.
pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/config.json ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "act",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.state": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 6
9
+ ]
10
+ },
11
+ "observation.images.front": {
12
+ "type": "VISUAL",
13
+ "shape": [
14
+ 3,
15
+ 480,
16
+ 640
17
+ ]
18
+ },
19
+ "observation.images.wrist": {
20
+ "type": "VISUAL",
21
+ "shape": [
22
+ 3,
23
+ 480,
24
+ 640
25
+ ]
26
+ }
27
+ },
28
+ "output_features": {
29
+ "action": {
30
+ "type": "ACTION",
31
+ "shape": [
32
+ 6
33
+ ]
34
+ }
35
+ },
36
+ "device": "cuda",
37
+ "use_amp": false,
38
+ "use_peft": false,
39
+ "push_to_hub": false,
40
+ "repo_id": null,
41
+ "private": null,
42
+ "tags": null,
43
+ "license": null,
44
+ "pretrained_path": null,
45
+ "chunk_size": 90,
46
+ "n_action_steps": 30,
47
+ "normalization_mapping": {
48
+ "VISUAL": "MEAN_STD",
49
+ "STATE": "MEAN_STD",
50
+ "ACTION": "MEAN_STD"
51
+ },
52
+ "vision_backbone": "resnet18",
53
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
54
+ "replace_final_stride_with_dilation": false,
55
+ "pre_norm": false,
56
+ "dim_model": 512,
57
+ "n_heads": 8,
58
+ "dim_feedforward": 3200,
59
+ "feedforward_activation": "relu",
60
+ "n_encoder_layers": 4,
61
+ "n_decoder_layers": 1,
62
+ "use_vae": true,
63
+ "latent_dim": 32,
64
+ "n_vae_encoder_layers": 4,
65
+ "temporal_ensemble_coeff": null,
66
+ "dropout": 0.1,
67
+ "kl_weight": 10.0,
68
+ "optimizer_lr": 1e-05,
69
+ "optimizer_weight_decay": 0.0001,
70
+ "optimizer_lr_backbone": 1e-05
71
+ }
pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0d0869a1169616e4daf5ba36c82019151ce66f09a659ed8fbd3c32bfb2201346
3
+ size 206658768
pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/policy_postprocessor.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "unnormalizer_processor",
6
+ "config": {
7
+ "eps": 1e-08,
8
+ "features": {
9
+ "action": {
10
+ "type": "ACTION",
11
+ "shape": [
12
+ 6
13
+ ]
14
+ }
15
+ },
16
+ "norm_map": {
17
+ "VISUAL": "MEAN_STD",
18
+ "STATE": "MEAN_STD",
19
+ "ACTION": "MEAN_STD"
20
+ }
21
+ },
22
+ "state_file": "policy_postprocessor_step_0_unnormalizer_processor.safetensors"
23
+ },
24
+ {
25
+ "registry_name": "device_processor",
26
+ "config": {
27
+ "device": "cpu",
28
+ "float_dtype": null
29
+ }
30
+ }
31
+ ]
32
+ }
pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:931ed929c95ef00f6e334882511164d4c9d89f16bfb31eeb212ddecf99897ed7
3
+ size 7552
pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/policy_preprocessor.json ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {}
8
+ }
9
+ },
10
+ {
11
+ "registry_name": "to_batch_processor",
12
+ "config": {}
13
+ },
14
+ {
15
+ "registry_name": "device_processor",
16
+ "config": {
17
+ "device": "cuda",
18
+ "float_dtype": null
19
+ }
20
+ },
21
+ {
22
+ "registry_name": "normalizer_processor",
23
+ "config": {
24
+ "eps": 1e-08,
25
+ "features": {
26
+ "observation.state": {
27
+ "type": "STATE",
28
+ "shape": [
29
+ 6
30
+ ]
31
+ },
32
+ "observation.images.front": {
33
+ "type": "VISUAL",
34
+ "shape": [
35
+ 3,
36
+ 480,
37
+ 640
38
+ ]
39
+ },
40
+ "observation.images.wrist": {
41
+ "type": "VISUAL",
42
+ "shape": [
43
+ 3,
44
+ 480,
45
+ 640
46
+ ]
47
+ },
48
+ "action": {
49
+ "type": "ACTION",
50
+ "shape": [
51
+ 6
52
+ ]
53
+ }
54
+ },
55
+ "norm_map": {
56
+ "VISUAL": "MEAN_STD",
57
+ "STATE": "MEAN_STD",
58
+ "ACTION": "MEAN_STD"
59
+ }
60
+ },
61
+ "state_file": "policy_preprocessor_step_3_normalizer_processor.safetensors"
62
+ }
63
+ ]
64
+ }
pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:931ed929c95ef00f6e334882511164d4c9d89f16bfb31eeb212ddecf99897ed7
3
+ size 7552
pick_v3_1-2-pieces/checkpoints/012000/pretrained_model/train_config.json ADDED
@@ -0,0 +1,533 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "local/so101_act",
4
+ "root": "C:\\Users\\Usuario\\Documents\\Workspace\\LeRobot_\\demos\\06_imitation_learning\\data\\real_so101\\act_pick_super",
5
+ "episodes": [
6
+ 64,
7
+ 65,
8
+ 70,
9
+ 74,
10
+ 78,
11
+ 79,
12
+ 84,
13
+ 89,
14
+ 90,
15
+ 93,
16
+ 94,
17
+ 100,
18
+ 101,
19
+ 102,
20
+ 103,
21
+ 104,
22
+ 105,
23
+ 106,
24
+ 107,
25
+ 109,
26
+ 110,
27
+ 111,
28
+ 112,
29
+ 113,
30
+ 114,
31
+ 115,
32
+ 116,
33
+ 117,
34
+ 118,
35
+ 119,
36
+ 120,
37
+ 121,
38
+ 122,
39
+ 124,
40
+ 125,
41
+ 126,
42
+ 127,
43
+ 128,
44
+ 129,
45
+ 130,
46
+ 131,
47
+ 132,
48
+ 133,
49
+ 134,
50
+ 135,
51
+ 136,
52
+ 137,
53
+ 138,
54
+ 139,
55
+ 140,
56
+ 141,
57
+ 142,
58
+ 143,
59
+ 144,
60
+ 145,
61
+ 146,
62
+ 147,
63
+ 148,
64
+ 149,
65
+ 150,
66
+ 151,
67
+ 152,
68
+ 153,
69
+ 154,
70
+ 155,
71
+ 156,
72
+ 157,
73
+ 158,
74
+ 159,
75
+ 160,
76
+ 161,
77
+ 162,
78
+ 163,
79
+ 164,
80
+ 165,
81
+ 166,
82
+ 167,
83
+ 168,
84
+ 169,
85
+ 170,
86
+ 171,
87
+ 172,
88
+ 173,
89
+ 174,
90
+ 175,
91
+ 176,
92
+ 177,
93
+ 178,
94
+ 179,
95
+ 180,
96
+ 181,
97
+ 182,
98
+ 183,
99
+ 184,
100
+ 185,
101
+ 186,
102
+ 187,
103
+ 188,
104
+ 189,
105
+ 190,
106
+ 191,
107
+ 192,
108
+ 193,
109
+ 194,
110
+ 195,
111
+ 196,
112
+ 197,
113
+ 198,
114
+ 199,
115
+ 200,
116
+ 201,
117
+ 202,
118
+ 203,
119
+ 204,
120
+ 205,
121
+ 206,
122
+ 207,
123
+ 208,
124
+ 209,
125
+ 210,
126
+ 211,
127
+ 212,
128
+ 213,
129
+ 214,
130
+ 215,
131
+ 216,
132
+ 217,
133
+ 218,
134
+ 219,
135
+ 220,
136
+ 221,
137
+ 222,
138
+ 223,
139
+ 224,
140
+ 225,
141
+ 226,
142
+ 227,
143
+ 228,
144
+ 229,
145
+ 230,
146
+ 231,
147
+ 232,
148
+ 233,
149
+ 235,
150
+ 236,
151
+ 237,
152
+ 238,
153
+ 239,
154
+ 240,
155
+ 241,
156
+ 242,
157
+ 243,
158
+ 244,
159
+ 245,
160
+ 246,
161
+ 247,
162
+ 248,
163
+ 249,
164
+ 250,
165
+ 251,
166
+ 252,
167
+ 253,
168
+ 255,
169
+ 256,
170
+ 257,
171
+ 258,
172
+ 259,
173
+ 260,
174
+ 261,
175
+ 262,
176
+ 263,
177
+ 264,
178
+ 265,
179
+ 266,
180
+ 267,
181
+ 268,
182
+ 269,
183
+ 270,
184
+ 271,
185
+ 272,
186
+ 273,
187
+ 274,
188
+ 275,
189
+ 276,
190
+ 277,
191
+ 278,
192
+ 279,
193
+ 280,
194
+ 281,
195
+ 282,
196
+ 283,
197
+ 284,
198
+ 285,
199
+ 286,
200
+ 287,
201
+ 288,
202
+ 289,
203
+ 290,
204
+ 291,
205
+ 292,
206
+ 293,
207
+ 294,
208
+ 295,
209
+ 296,
210
+ 297,
211
+ 298,
212
+ 299,
213
+ 300,
214
+ 301,
215
+ 302,
216
+ 303,
217
+ 304,
218
+ 305,
219
+ 306,
220
+ 307,
221
+ 308,
222
+ 309,
223
+ 310,
224
+ 311,
225
+ 312,
226
+ 313,
227
+ 314,
228
+ 315,
229
+ 316,
230
+ 317,
231
+ 318,
232
+ 319,
233
+ 320,
234
+ 325,
235
+ 326,
236
+ 327,
237
+ 329,
238
+ 330,
239
+ 331,
240
+ 332,
241
+ 333,
242
+ 334,
243
+ 335,
244
+ 336,
245
+ 337,
246
+ 338,
247
+ 339,
248
+ 340,
249
+ 341,
250
+ 342,
251
+ 343,
252
+ 344,
253
+ 345,
254
+ 346,
255
+ 347,
256
+ 348,
257
+ 349,
258
+ 351,
259
+ 352,
260
+ 353,
261
+ 354,
262
+ 355,
263
+ 356,
264
+ 357,
265
+ 358,
266
+ 359,
267
+ 360,
268
+ 361,
269
+ 364,
270
+ 365,
271
+ 366,
272
+ 367,
273
+ 368,
274
+ 369,
275
+ 370,
276
+ 371,
277
+ 372,
278
+ 373,
279
+ 374,
280
+ 375,
281
+ 376,
282
+ 377,
283
+ 378,
284
+ 391,
285
+ 392,
286
+ 393,
287
+ 394,
288
+ 395,
289
+ 396,
290
+ 397,
291
+ 398,
292
+ 399,
293
+ 400,
294
+ 401,
295
+ 402,
296
+ 403,
297
+ 404,
298
+ 405,
299
+ 406,
300
+ 407,
301
+ 408,
302
+ 409,
303
+ 410,
304
+ 411,
305
+ 412,
306
+ 413,
307
+ 414,
308
+ 415,
309
+ 416,
310
+ 417,
311
+ 419,
312
+ 420,
313
+ 421,
314
+ 422,
315
+ 423,
316
+ 424,
317
+ 425,
318
+ 426,
319
+ 427,
320
+ 428,
321
+ 429,
322
+ 430,
323
+ 431,
324
+ 432,
325
+ 433,
326
+ 434,
327
+ 435,
328
+ 436,
329
+ 438,
330
+ 439,
331
+ 440,
332
+ 441,
333
+ 442,
334
+ 443,
335
+ 444,
336
+ 445
337
+ ],
338
+ "image_transforms": {
339
+ "enable": true,
340
+ "max_num_transforms": 3,
341
+ "random_order": false,
342
+ "tfs": {
343
+ "brightness": {
344
+ "weight": 1.0,
345
+ "type": "ColorJitter",
346
+ "kwargs": {
347
+ "brightness": [
348
+ 0.8,
349
+ 1.2
350
+ ]
351
+ }
352
+ },
353
+ "contrast": {
354
+ "weight": 1.0,
355
+ "type": "ColorJitter",
356
+ "kwargs": {
357
+ "contrast": [
358
+ 0.8,
359
+ 1.2
360
+ ]
361
+ }
362
+ },
363
+ "saturation": {
364
+ "weight": 1.0,
365
+ "type": "ColorJitter",
366
+ "kwargs": {
367
+ "saturation": [
368
+ 0.5,
369
+ 1.5
370
+ ]
371
+ }
372
+ },
373
+ "hue": {
374
+ "weight": 1.0,
375
+ "type": "ColorJitter",
376
+ "kwargs": {
377
+ "hue": [
378
+ -0.05,
379
+ 0.05
380
+ ]
381
+ }
382
+ },
383
+ "sharpness": {
384
+ "weight": 1.0,
385
+ "type": "SharpnessJitter",
386
+ "kwargs": {
387
+ "sharpness": [
388
+ 0.5,
389
+ 1.5
390
+ ]
391
+ }
392
+ },
393
+ "affine": {
394
+ "weight": 1.0,
395
+ "type": "RandomAffine",
396
+ "kwargs": {
397
+ "degrees": [
398
+ -5.0,
399
+ 5.0
400
+ ],
401
+ "translate": [
402
+ 0.05,
403
+ 0.05
404
+ ]
405
+ }
406
+ }
407
+ }
408
+ },
409
+ "revision": null,
410
+ "use_imagenet_stats": true,
411
+ "video_backend": "pyav",
412
+ "streaming": false
413
+ },
414
+ "env": null,
415
+ "policy": {
416
+ "type": "act",
417
+ "n_obs_steps": 1,
418
+ "input_features": {
419
+ "observation.state": {
420
+ "type": "STATE",
421
+ "shape": [
422
+ 6
423
+ ]
424
+ },
425
+ "observation.images.front": {
426
+ "type": "VISUAL",
427
+ "shape": [
428
+ 3,
429
+ 480,
430
+ 640
431
+ ]
432
+ },
433
+ "observation.images.wrist": {
434
+ "type": "VISUAL",
435
+ "shape": [
436
+ 3,
437
+ 480,
438
+ 640
439
+ ]
440
+ }
441
+ },
442
+ "output_features": {
443
+ "action": {
444
+ "type": "ACTION",
445
+ "shape": [
446
+ 6
447
+ ]
448
+ }
449
+ },
450
+ "device": "cuda",
451
+ "use_amp": false,
452
+ "use_peft": false,
453
+ "push_to_hub": false,
454
+ "repo_id": null,
455
+ "private": null,
456
+ "tags": null,
457
+ "license": null,
458
+ "pretrained_path": null,
459
+ "chunk_size": 90,
460
+ "n_action_steps": 30,
461
+ "normalization_mapping": {
462
+ "VISUAL": "MEAN_STD",
463
+ "STATE": "MEAN_STD",
464
+ "ACTION": "MEAN_STD"
465
+ },
466
+ "vision_backbone": "resnet18",
467
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
468
+ "replace_final_stride_with_dilation": false,
469
+ "pre_norm": false,
470
+ "dim_model": 512,
471
+ "n_heads": 8,
472
+ "dim_feedforward": 3200,
473
+ "feedforward_activation": "relu",
474
+ "n_encoder_layers": 4,
475
+ "n_decoder_layers": 1,
476
+ "use_vae": true,
477
+ "latent_dim": 32,
478
+ "n_vae_encoder_layers": 4,
479
+ "temporal_ensemble_coeff": null,
480
+ "dropout": 0.1,
481
+ "kl_weight": 10.0,
482
+ "optimizer_lr": 1e-05,
483
+ "optimizer_weight_decay": 0.0001,
484
+ "optimizer_lr_backbone": 1e-05
485
+ },
486
+ "output_dir": "C:\\Users\\Usuario\\Documents\\Workspace\\LeRobot_\\demos\\06_imitation_learning\\outputs\\sweep_v3",
487
+ "job_name": "act_so101_sweep_v3",
488
+ "resume": false,
489
+ "seed": 1000,
490
+ "num_workers": 4,
491
+ "batch_size": 8,
492
+ "steps": 30000,
493
+ "eval_freq": 20000,
494
+ "log_freq": 100,
495
+ "tolerance_s": 0.0001,
496
+ "save_checkpoint": true,
497
+ "save_freq": 6000,
498
+ "use_policy_training_preset": true,
499
+ "optimizer": {
500
+ "type": "adamw",
501
+ "lr": 1e-05,
502
+ "weight_decay": 0.0001,
503
+ "grad_clip_norm": 10.0,
504
+ "betas": [
505
+ 0.9,
506
+ 0.999
507
+ ],
508
+ "eps": 1e-08
509
+ },
510
+ "scheduler": null,
511
+ "eval": {
512
+ "n_episodes": 50,
513
+ "batch_size": 50,
514
+ "use_async_envs": false
515
+ },
516
+ "wandb": {
517
+ "enable": false,
518
+ "disable_artifact": false,
519
+ "project": "lerobot",
520
+ "entity": null,
521
+ "notes": null,
522
+ "run_id": null,
523
+ "mode": null
524
+ },
525
+ "peft": null,
526
+ "use_rabc": false,
527
+ "rabc_progress_path": null,
528
+ "rabc_kappa": 0.01,
529
+ "rabc_epsilon": 1e-06,
530
+ "rabc_head_mode": "sparse",
531
+ "rename_map": {},
532
+ "checkpoint_path": null
533
+ }
pick_v3_1-2-pieces/checkpoints/018000/dataset_features.json ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action": {
3
+ "dtype": "float32",
4
+ "names": [
5
+ "shoulder_pan.pos",
6
+ "shoulder_lift.pos",
7
+ "elbow_flex.pos",
8
+ "wrist_flex.pos",
9
+ "wrist_roll.pos",
10
+ "gripper.pos"
11
+ ],
12
+ "shape": [
13
+ 6
14
+ ]
15
+ },
16
+ "observation.images.front": {
17
+ "dtype": "video",
18
+ "shape": [
19
+ 480,
20
+ 640,
21
+ 3
22
+ ],
23
+ "names": [
24
+ "height",
25
+ "width",
26
+ "channels"
27
+ ],
28
+ "info": {
29
+ "is_depth_map": false,
30
+ "video.height": 480,
31
+ "video.width": 640,
32
+ "video.codec": "h264",
33
+ "video.pix_fmt": "yuv420p",
34
+ "video.fps": 30,
35
+ "video.channels": 3,
36
+ "has_audio": false,
37
+ "video.g": 2,
38
+ "video.crf": 30,
39
+ "video.preset": "p1",
40
+ "video.fast_decode": 0,
41
+ "video.video_backend": "pyav",
42
+ "video.extra_options": {}
43
+ }
44
+ },
45
+ "observation.images.wrist": {
46
+ "dtype": "video",
47
+ "shape": [
48
+ 480,
49
+ 640,
50
+ 3
51
+ ],
52
+ "names": [
53
+ "height",
54
+ "width",
55
+ "channels"
56
+ ],
57
+ "info": {
58
+ "is_depth_map": false,
59
+ "video.height": 480,
60
+ "video.width": 640,
61
+ "video.codec": "h264",
62
+ "video.pix_fmt": "yuv420p",
63
+ "video.fps": 30,
64
+ "video.channels": 3,
65
+ "has_audio": false,
66
+ "video.g": 2,
67
+ "video.crf": 30,
68
+ "video.preset": "p1",
69
+ "video.fast_decode": 0,
70
+ "video.video_backend": "pyav",
71
+ "video.extra_options": {}
72
+ }
73
+ },
74
+ "observation.state": {
75
+ "dtype": "float32",
76
+ "names": [
77
+ "shoulder_pan.pos",
78
+ "shoulder_lift.pos",
79
+ "elbow_flex.pos",
80
+ "wrist_flex.pos",
81
+ "wrist_roll.pos",
82
+ "gripper.pos"
83
+ ],
84
+ "shape": [
85
+ 6
86
+ ]
87
+ }
88
+ }
pick_v3_1-2-pieces/checkpoints/018000/model_card_original.md ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: lerobot
4
+ pipeline_tag: robotics
5
+ tags:
6
+ - lerobot
7
+ - act
8
+ - so101
9
+ - pick
10
+ ---
11
+
12
+ # act-pick-v3-1-2-pieces-18000
13
+
14
+ The v2 recipe plus the two new evening sessions of 24 Oct (third round, v3): checkpoint at 18,000 steps of a 30,000-step run.
15
+
16
+ ACT policy for the SO-101 "pick a piece from the crate and hold it up to the camera" task. Inputs: front camera, wrist camera and the six joint positions; output: a chunk of 90 future joint targets at 30 fps. This is the checkpoint at 18,000 steps.
17
+
18
+ ## Training data
19
+
20
+ 331 training episodes, 30 fps only, one or two pieces in the crate at the start (every 24 Oct session, plus 12 episodes of 23 Oct), 5 to 15 s long, start gap at most 20 degrees, at most 3 s idle at either end, gripper opens at least once, known bad takes removed.
21
+
22
+ Recipe: LeRobot `train()`, batch 8, chunk 90, 30 action steps per replan, learning rate 1e-5, LeRobot default image augmentation (colour, sharpness, small affine), seed 1000.
23
+
24
+ ## Offline check on the held-out episodes
25
+
26
+ Held out: 16 episodes, 12 from `so101-pick-cube_20261024_181939` (one piece, varied positions) and 4 from `so101-pick-cube_20261024_144537` (two pieces). Mean absolute joint error over the predicted chunk (lower is better) against a "stay at the current joint positions" baseline of 20.71 (16.23 for chunk 60, which looks a shorter way into the future, so its raw error is not comparable with the others):
27
+
28
+ | run | training episodes | best checkpoint | error | below stay-still | repo |
29
+ |---|---|---|---|---|---|
30
+ | super (all quality-passing data) | 398 | 6k | 15.16 | 27% | act-pick-v3-6000 |
31
+ | strict (6-12 s, tight quality) | 205 | 2.4k | 15.14 | 27% | act-pick-v3-strict-2400 |
32
+ | v3 recipe on 1-2 piece episodes | 331 | 18k | 15.35 | 26% | act-pick-v3-1-2-pieces-18000 |
33
+ | v2 (earlier) | 278 | 6k | 15.36 | 26% | act-pick-v2-6000 |
34
+ | chunk 60 (own baseline 16.23) | 331 | 4.8k | 11.94 | 26% | act-pick-v3-chunk60-4800 |
35
+ | recent sessions only | 100 | 12k | 16.87 | 19% | not uploaded |
36
+ | first run | 165 | 6k | 16.87 | 19% | act-pick-* |
37
+
38
+ This model: pooled error 15.35 (26% below stay-still); per joint, policy / baseline: shoulder_pan 7.2 / 9.7, shoulder_lift 13.6 / 30.2, elbow_flex 15.1 / 21.0, wrist_flex 15.1 / 20.4, wrist_roll 33.4 / 28.5, gripper 7.7 / 14.5. Per session: 15.83 on the one-piece session, 13.94 on the two-piece session. Other checkpoints of this run: 15.86 (6k), 15.89 (12k), 15.73 (24k), 16.38 (30k).
39
+
40
+ The best checkpoints of the first five runs are within noise of each other, and every run got worse after about one pass over the data. The offline error cannot say which model grasps best; none of them has been tested on the robot.
pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/config.json ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "act",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.state": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 6
9
+ ]
10
+ },
11
+ "observation.images.front": {
12
+ "type": "VISUAL",
13
+ "shape": [
14
+ 3,
15
+ 480,
16
+ 640
17
+ ]
18
+ },
19
+ "observation.images.wrist": {
20
+ "type": "VISUAL",
21
+ "shape": [
22
+ 3,
23
+ 480,
24
+ 640
25
+ ]
26
+ }
27
+ },
28
+ "output_features": {
29
+ "action": {
30
+ "type": "ACTION",
31
+ "shape": [
32
+ 6
33
+ ]
34
+ }
35
+ },
36
+ "device": "cuda",
37
+ "use_amp": false,
38
+ "use_peft": false,
39
+ "push_to_hub": false,
40
+ "repo_id": null,
41
+ "private": null,
42
+ "tags": null,
43
+ "license": null,
44
+ "pretrained_path": null,
45
+ "chunk_size": 90,
46
+ "n_action_steps": 30,
47
+ "normalization_mapping": {
48
+ "VISUAL": "MEAN_STD",
49
+ "STATE": "MEAN_STD",
50
+ "ACTION": "MEAN_STD"
51
+ },
52
+ "vision_backbone": "resnet18",
53
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
54
+ "replace_final_stride_with_dilation": false,
55
+ "pre_norm": false,
56
+ "dim_model": 512,
57
+ "n_heads": 8,
58
+ "dim_feedforward": 3200,
59
+ "feedforward_activation": "relu",
60
+ "n_encoder_layers": 4,
61
+ "n_decoder_layers": 1,
62
+ "use_vae": true,
63
+ "latent_dim": 32,
64
+ "n_vae_encoder_layers": 4,
65
+ "temporal_ensemble_coeff": null,
66
+ "dropout": 0.1,
67
+ "kl_weight": 10.0,
68
+ "optimizer_lr": 1e-05,
69
+ "optimizer_weight_decay": 0.0001,
70
+ "optimizer_lr_backbone": 1e-05
71
+ }
pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:922759479f1ae8bb32b1a83e9fe0c7667584d362e1d6d39eac09bc907a53525a
3
+ size 206658768
pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/policy_postprocessor.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "unnormalizer_processor",
6
+ "config": {
7
+ "eps": 1e-08,
8
+ "features": {
9
+ "action": {
10
+ "type": "ACTION",
11
+ "shape": [
12
+ 6
13
+ ]
14
+ }
15
+ },
16
+ "norm_map": {
17
+ "VISUAL": "MEAN_STD",
18
+ "STATE": "MEAN_STD",
19
+ "ACTION": "MEAN_STD"
20
+ }
21
+ },
22
+ "state_file": "policy_postprocessor_step_0_unnormalizer_processor.safetensors"
23
+ },
24
+ {
25
+ "registry_name": "device_processor",
26
+ "config": {
27
+ "device": "cpu",
28
+ "float_dtype": null
29
+ }
30
+ }
31
+ ]
32
+ }
pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:931ed929c95ef00f6e334882511164d4c9d89f16bfb31eeb212ddecf99897ed7
3
+ size 7552
pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/policy_preprocessor.json ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {}
8
+ }
9
+ },
10
+ {
11
+ "registry_name": "to_batch_processor",
12
+ "config": {}
13
+ },
14
+ {
15
+ "registry_name": "device_processor",
16
+ "config": {
17
+ "device": "cuda",
18
+ "float_dtype": null
19
+ }
20
+ },
21
+ {
22
+ "registry_name": "normalizer_processor",
23
+ "config": {
24
+ "eps": 1e-08,
25
+ "features": {
26
+ "observation.state": {
27
+ "type": "STATE",
28
+ "shape": [
29
+ 6
30
+ ]
31
+ },
32
+ "observation.images.front": {
33
+ "type": "VISUAL",
34
+ "shape": [
35
+ 3,
36
+ 480,
37
+ 640
38
+ ]
39
+ },
40
+ "observation.images.wrist": {
41
+ "type": "VISUAL",
42
+ "shape": [
43
+ 3,
44
+ 480,
45
+ 640
46
+ ]
47
+ },
48
+ "action": {
49
+ "type": "ACTION",
50
+ "shape": [
51
+ 6
52
+ ]
53
+ }
54
+ },
55
+ "norm_map": {
56
+ "VISUAL": "MEAN_STD",
57
+ "STATE": "MEAN_STD",
58
+ "ACTION": "MEAN_STD"
59
+ }
60
+ },
61
+ "state_file": "policy_preprocessor_step_3_normalizer_processor.safetensors"
62
+ }
63
+ ]
64
+ }
pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:931ed929c95ef00f6e334882511164d4c9d89f16bfb31eeb212ddecf99897ed7
3
+ size 7552
pick_v3_1-2-pieces/checkpoints/018000/pretrained_model/train_config.json ADDED
@@ -0,0 +1,533 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "local/so101_act",
4
+ "root": "C:\\Users\\Usuario\\Documents\\Workspace\\LeRobot_\\demos\\06_imitation_learning\\data\\real_so101\\act_pick_super",
5
+ "episodes": [
6
+ 64,
7
+ 65,
8
+ 70,
9
+ 74,
10
+ 78,
11
+ 79,
12
+ 84,
13
+ 89,
14
+ 90,
15
+ 93,
16
+ 94,
17
+ 100,
18
+ 101,
19
+ 102,
20
+ 103,
21
+ 104,
22
+ 105,
23
+ 106,
24
+ 107,
25
+ 109,
26
+ 110,
27
+ 111,
28
+ 112,
29
+ 113,
30
+ 114,
31
+ 115,
32
+ 116,
33
+ 117,
34
+ 118,
35
+ 119,
36
+ 120,
37
+ 121,
38
+ 122,
39
+ 124,
40
+ 125,
41
+ 126,
42
+ 127,
43
+ 128,
44
+ 129,
45
+ 130,
46
+ 131,
47
+ 132,
48
+ 133,
49
+ 134,
50
+ 135,
51
+ 136,
52
+ 137,
53
+ 138,
54
+ 139,
55
+ 140,
56
+ 141,
57
+ 142,
58
+ 143,
59
+ 144,
60
+ 145,
61
+ 146,
62
+ 147,
63
+ 148,
64
+ 149,
65
+ 150,
66
+ 151,
67
+ 152,
68
+ 153,
69
+ 154,
70
+ 155,
71
+ 156,
72
+ 157,
73
+ 158,
74
+ 159,
75
+ 160,
76
+ 161,
77
+ 162,
78
+ 163,
79
+ 164,
80
+ 165,
81
+ 166,
82
+ 167,
83
+ 168,
84
+ 169,
85
+ 170,
86
+ 171,
87
+ 172,
88
+ 173,
89
+ 174,
90
+ 175,
91
+ 176,
92
+ 177,
93
+ 178,
94
+ 179,
95
+ 180,
96
+ 181,
97
+ 182,
98
+ 183,
99
+ 184,
100
+ 185,
101
+ 186,
102
+ 187,
103
+ 188,
104
+ 189,
105
+ 190,
106
+ 191,
107
+ 192,
108
+ 193,
109
+ 194,
110
+ 195,
111
+ 196,
112
+ 197,
113
+ 198,
114
+ 199,
115
+ 200,
116
+ 201,
117
+ 202,
118
+ 203,
119
+ 204,
120
+ 205,
121
+ 206,
122
+ 207,
123
+ 208,
124
+ 209,
125
+ 210,
126
+ 211,
127
+ 212,
128
+ 213,
129
+ 214,
130
+ 215,
131
+ 216,
132
+ 217,
133
+ 218,
134
+ 219,
135
+ 220,
136
+ 221,
137
+ 222,
138
+ 223,
139
+ 224,
140
+ 225,
141
+ 226,
142
+ 227,
143
+ 228,
144
+ 229,
145
+ 230,
146
+ 231,
147
+ 232,
148
+ 233,
149
+ 235,
150
+ 236,
151
+ 237,
152
+ 238,
153
+ 239,
154
+ 240,
155
+ 241,
156
+ 242,
157
+ 243,
158
+ 244,
159
+ 245,
160
+ 246,
161
+ 247,
162
+ 248,
163
+ 249,
164
+ 250,
165
+ 251,
166
+ 252,
167
+ 253,
168
+ 255,
169
+ 256,
170
+ 257,
171
+ 258,
172
+ 259,
173
+ 260,
174
+ 261,
175
+ 262,
176
+ 263,
177
+ 264,
178
+ 265,
179
+ 266,
180
+ 267,
181
+ 268,
182
+ 269,
183
+ 270,
184
+ 271,
185
+ 272,
186
+ 273,
187
+ 274,
188
+ 275,
189
+ 276,
190
+ 277,
191
+ 278,
192
+ 279,
193
+ 280,
194
+ 281,
195
+ 282,
196
+ 283,
197
+ 284,
198
+ 285,
199
+ 286,
200
+ 287,
201
+ 288,
202
+ 289,
203
+ 290,
204
+ 291,
205
+ 292,
206
+ 293,
207
+ 294,
208
+ 295,
209
+ 296,
210
+ 297,
211
+ 298,
212
+ 299,
213
+ 300,
214
+ 301,
215
+ 302,
216
+ 303,
217
+ 304,
218
+ 305,
219
+ 306,
220
+ 307,
221
+ 308,
222
+ 309,
223
+ 310,
224
+ 311,
225
+ 312,
226
+ 313,
227
+ 314,
228
+ 315,
229
+ 316,
230
+ 317,
231
+ 318,
232
+ 319,
233
+ 320,
234
+ 325,
235
+ 326,
236
+ 327,
237
+ 329,
238
+ 330,
239
+ 331,
240
+ 332,
241
+ 333,
242
+ 334,
243
+ 335,
244
+ 336,
245
+ 337,
246
+ 338,
247
+ 339,
248
+ 340,
249
+ 341,
250
+ 342,
251
+ 343,
252
+ 344,
253
+ 345,
254
+ 346,
255
+ 347,
256
+ 348,
257
+ 349,
258
+ 351,
259
+ 352,
260
+ 353,
261
+ 354,
262
+ 355,
263
+ 356,
264
+ 357,
265
+ 358,
266
+ 359,
267
+ 360,
268
+ 361,
269
+ 364,
270
+ 365,
271
+ 366,
272
+ 367,
273
+ 368,
274
+ 369,
275
+ 370,
276
+ 371,
277
+ 372,
278
+ 373,
279
+ 374,
280
+ 375,
281
+ 376,
282
+ 377,
283
+ 378,
284
+ 391,
285
+ 392,
286
+ 393,
287
+ 394,
288
+ 395,
289
+ 396,
290
+ 397,
291
+ 398,
292
+ 399,
293
+ 400,
294
+ 401,
295
+ 402,
296
+ 403,
297
+ 404,
298
+ 405,
299
+ 406,
300
+ 407,
301
+ 408,
302
+ 409,
303
+ 410,
304
+ 411,
305
+ 412,
306
+ 413,
307
+ 414,
308
+ 415,
309
+ 416,
310
+ 417,
311
+ 419,
312
+ 420,
313
+ 421,
314
+ 422,
315
+ 423,
316
+ 424,
317
+ 425,
318
+ 426,
319
+ 427,
320
+ 428,
321
+ 429,
322
+ 430,
323
+ 431,
324
+ 432,
325
+ 433,
326
+ 434,
327
+ 435,
328
+ 436,
329
+ 438,
330
+ 439,
331
+ 440,
332
+ 441,
333
+ 442,
334
+ 443,
335
+ 444,
336
+ 445
337
+ ],
338
+ "image_transforms": {
339
+ "enable": true,
340
+ "max_num_transforms": 3,
341
+ "random_order": false,
342
+ "tfs": {
343
+ "brightness": {
344
+ "weight": 1.0,
345
+ "type": "ColorJitter",
346
+ "kwargs": {
347
+ "brightness": [
348
+ 0.8,
349
+ 1.2
350
+ ]
351
+ }
352
+ },
353
+ "contrast": {
354
+ "weight": 1.0,
355
+ "type": "ColorJitter",
356
+ "kwargs": {
357
+ "contrast": [
358
+ 0.8,
359
+ 1.2
360
+ ]
361
+ }
362
+ },
363
+ "saturation": {
364
+ "weight": 1.0,
365
+ "type": "ColorJitter",
366
+ "kwargs": {
367
+ "saturation": [
368
+ 0.5,
369
+ 1.5
370
+ ]
371
+ }
372
+ },
373
+ "hue": {
374
+ "weight": 1.0,
375
+ "type": "ColorJitter",
376
+ "kwargs": {
377
+ "hue": [
378
+ -0.05,
379
+ 0.05
380
+ ]
381
+ }
382
+ },
383
+ "sharpness": {
384
+ "weight": 1.0,
385
+ "type": "SharpnessJitter",
386
+ "kwargs": {
387
+ "sharpness": [
388
+ 0.5,
389
+ 1.5
390
+ ]
391
+ }
392
+ },
393
+ "affine": {
394
+ "weight": 1.0,
395
+ "type": "RandomAffine",
396
+ "kwargs": {
397
+ "degrees": [
398
+ -5.0,
399
+ 5.0
400
+ ],
401
+ "translate": [
402
+ 0.05,
403
+ 0.05
404
+ ]
405
+ }
406
+ }
407
+ }
408
+ },
409
+ "revision": null,
410
+ "use_imagenet_stats": true,
411
+ "video_backend": "pyav",
412
+ "streaming": false
413
+ },
414
+ "env": null,
415
+ "policy": {
416
+ "type": "act",
417
+ "n_obs_steps": 1,
418
+ "input_features": {
419
+ "observation.state": {
420
+ "type": "STATE",
421
+ "shape": [
422
+ 6
423
+ ]
424
+ },
425
+ "observation.images.front": {
426
+ "type": "VISUAL",
427
+ "shape": [
428
+ 3,
429
+ 480,
430
+ 640
431
+ ]
432
+ },
433
+ "observation.images.wrist": {
434
+ "type": "VISUAL",
435
+ "shape": [
436
+ 3,
437
+ 480,
438
+ 640
439
+ ]
440
+ }
441
+ },
442
+ "output_features": {
443
+ "action": {
444
+ "type": "ACTION",
445
+ "shape": [
446
+ 6
447
+ ]
448
+ }
449
+ },
450
+ "device": "cuda",
451
+ "use_amp": false,
452
+ "use_peft": false,
453
+ "push_to_hub": false,
454
+ "repo_id": null,
455
+ "private": null,
456
+ "tags": null,
457
+ "license": null,
458
+ "pretrained_path": null,
459
+ "chunk_size": 90,
460
+ "n_action_steps": 30,
461
+ "normalization_mapping": {
462
+ "VISUAL": "MEAN_STD",
463
+ "STATE": "MEAN_STD",
464
+ "ACTION": "MEAN_STD"
465
+ },
466
+ "vision_backbone": "resnet18",
467
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
468
+ "replace_final_stride_with_dilation": false,
469
+ "pre_norm": false,
470
+ "dim_model": 512,
471
+ "n_heads": 8,
472
+ "dim_feedforward": 3200,
473
+ "feedforward_activation": "relu",
474
+ "n_encoder_layers": 4,
475
+ "n_decoder_layers": 1,
476
+ "use_vae": true,
477
+ "latent_dim": 32,
478
+ "n_vae_encoder_layers": 4,
479
+ "temporal_ensemble_coeff": null,
480
+ "dropout": 0.1,
481
+ "kl_weight": 10.0,
482
+ "optimizer_lr": 1e-05,
483
+ "optimizer_weight_decay": 0.0001,
484
+ "optimizer_lr_backbone": 1e-05
485
+ },
486
+ "output_dir": "C:\\Users\\Usuario\\Documents\\Workspace\\LeRobot_\\demos\\06_imitation_learning\\outputs\\sweep_v3",
487
+ "job_name": "act_so101_sweep_v3",
488
+ "resume": false,
489
+ "seed": 1000,
490
+ "num_workers": 4,
491
+ "batch_size": 8,
492
+ "steps": 30000,
493
+ "eval_freq": 20000,
494
+ "log_freq": 100,
495
+ "tolerance_s": 0.0001,
496
+ "save_checkpoint": true,
497
+ "save_freq": 6000,
498
+ "use_policy_training_preset": true,
499
+ "optimizer": {
500
+ "type": "adamw",
501
+ "lr": 1e-05,
502
+ "weight_decay": 0.0001,
503
+ "grad_clip_norm": 10.0,
504
+ "betas": [
505
+ 0.9,
506
+ 0.999
507
+ ],
508
+ "eps": 1e-08
509
+ },
510
+ "scheduler": null,
511
+ "eval": {
512
+ "n_episodes": 50,
513
+ "batch_size": 50,
514
+ "use_async_envs": false
515
+ },
516
+ "wandb": {
517
+ "enable": false,
518
+ "disable_artifact": false,
519
+ "project": "lerobot",
520
+ "entity": null,
521
+ "notes": null,
522
+ "run_id": null,
523
+ "mode": null
524
+ },
525
+ "peft": null,
526
+ "use_rabc": false,
527
+ "rabc_progress_path": null,
528
+ "rabc_kappa": 0.01,
529
+ "rabc_epsilon": 1e-06,
530
+ "rabc_head_mode": "sparse",
531
+ "rename_map": {},
532
+ "checkpoint_path": null
533
+ }
pick_v3_1-2-pieces/checkpoints/024000/dataset_features.json ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action": {
3
+ "dtype": "float32",
4
+ "names": [
5
+ "shoulder_pan.pos",
6
+ "shoulder_lift.pos",
7
+ "elbow_flex.pos",
8
+ "wrist_flex.pos",
9
+ "wrist_roll.pos",
10
+ "gripper.pos"
11
+ ],
12
+ "shape": [
13
+ 6
14
+ ]
15
+ },
16
+ "observation.images.front": {
17
+ "dtype": "video",
18
+ "shape": [
19
+ 480,
20
+ 640,
21
+ 3
22
+ ],
23
+ "names": [
24
+ "height",
25
+ "width",
26
+ "channels"
27
+ ],
28
+ "info": {
29
+ "is_depth_map": false,
30
+ "video.height": 480,
31
+ "video.width": 640,
32
+ "video.codec": "h264",
33
+ "video.pix_fmt": "yuv420p",
34
+ "video.fps": 30,
35
+ "video.channels": 3,
36
+ "has_audio": false,
37
+ "video.g": 2,
38
+ "video.crf": 30,
39
+ "video.preset": "p1",
40
+ "video.fast_decode": 0,
41
+ "video.video_backend": "pyav",
42
+ "video.extra_options": {}
43
+ }
44
+ },
45
+ "observation.images.wrist": {
46
+ "dtype": "video",
47
+ "shape": [
48
+ 480,
49
+ 640,
50
+ 3
51
+ ],
52
+ "names": [
53
+ "height",
54
+ "width",
55
+ "channels"
56
+ ],
57
+ "info": {
58
+ "is_depth_map": false,
59
+ "video.height": 480,
60
+ "video.width": 640,
61
+ "video.codec": "h264",
62
+ "video.pix_fmt": "yuv420p",
63
+ "video.fps": 30,
64
+ "video.channels": 3,
65
+ "has_audio": false,
66
+ "video.g": 2,
67
+ "video.crf": 30,
68
+ "video.preset": "p1",
69
+ "video.fast_decode": 0,
70
+ "video.video_backend": "pyav",
71
+ "video.extra_options": {}
72
+ }
73
+ },
74
+ "observation.state": {
75
+ "dtype": "float32",
76
+ "names": [
77
+ "shoulder_pan.pos",
78
+ "shoulder_lift.pos",
79
+ "elbow_flex.pos",
80
+ "wrist_flex.pos",
81
+ "wrist_roll.pos",
82
+ "gripper.pos"
83
+ ],
84
+ "shape": [
85
+ 6
86
+ ]
87
+ }
88
+ }
pick_v3_1-2-pieces/checkpoints/024000/model_card_original.md ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: lerobot
4
+ pipeline_tag: robotics
5
+ tags:
6
+ - lerobot
7
+ - act
8
+ - so101
9
+ - pick
10
+ ---
11
+
12
+ # act-pick-v3-1-2-pieces-24000
13
+
14
+ The v2 recipe plus the two new evening sessions of 24 Oct (third round, v3), 30,000-step run. Checkpoint at 24,000 steps. This is not the best checkpoint of its run (the best one is at 18,000 steps); it is published for completeness.
15
+
16
+ ACT policy for the SO-101 "pick a piece from the crate and hold it up to the camera" task. Inputs: front camera, wrist camera and the six joint positions; output: a chunk of 90 future joint targets at 30 fps.
17
+
18
+ ## Training data
19
+
20
+ 331 training episodes, 30 fps only, one or two pieces in the crate at the start (every 24 Oct session, plus 12 episodes of 23 Oct), 5 to 15 s long, start gap at most 20 degrees, at most 3 s idle at either end, gripper opens at least once, known bad takes removed.
21
+
22
+ Recipe: LeRobot `train()`, batch 8, chunk 90, 30 action steps per replan, default image augmentation, seed 1000.
23
+
24
+ ## Offline check on the held-out episodes
25
+
26
+ Held out: 16 episodes, 12 from `so101-pick-cube_20261024_181939` (one piece) and 4 from `so101-pick-cube_20261024_144537` (two pieces). Mean absolute joint error over the predicted chunk (lower is better) against a "stay at the current joint positions" baseline of 20.71:
27
+
28
+ | checkpoint of this run | error | below baseline |
29
+ |---|---|---|
30
+ | 6,000 steps | 15.86 | 23% |
31
+ | 12,000 steps | 15.89 | 23% |
32
+ | 18,000 steps | 15.35 | 26% |
33
+ | **24,000 steps (this one)** | **15.73** | 24% |
34
+ | 30,000 steps | 16.38 | 21% |
35
+
36
+ All runs of this round, best checkpoint of each:
37
+
38
+ | run | training episodes | best checkpoint | error | below stay-still | repo |
39
+ |---|---|---|---|---|---|
40
+ | super (all quality-passing data) | 398 | 6k | 15.16 | 27% | act-pick-v3-6000 |
41
+ | strict (6-12 s, tight quality) | 205 | 2.4k | 15.14 | 27% | act-pick-v3-strict-2400 |
42
+ | v3 recipe on 1-2 piece episodes | 331 | 18k | 15.35 | 26% | act-pick-v3-1-2-pieces-18000 |
43
+ | v2 (earlier) | 278 | 6k | 15.36 | 26% | act-pick-v2-6000 |
44
+ | chunk 60 (own baseline 16.23) | 331 | 4.8k | 11.94 | 26% | act-pick-v3-chunk60-4800 |
45
+ | one piece only | 292 | 9.6k | 15.47 | 25% | act-pick-v3-1piece-9600 |
46
+ | recent sessions only | 100 | 12k | 16.87 | 19% | act-pick-v3-recent-12000 |
47
+ | first run | 165 | 6k | 16.87 | 19% | act-pick-* |
48
+
49
+ The best checkpoints of the main runs are within noise of each other, and every run got worse after about one pass over the data. The offline error cannot say which model grasps best; none of them has been tested on the robot.
pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/config.json ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "act",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.state": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 6
9
+ ]
10
+ },
11
+ "observation.images.front": {
12
+ "type": "VISUAL",
13
+ "shape": [
14
+ 3,
15
+ 480,
16
+ 640
17
+ ]
18
+ },
19
+ "observation.images.wrist": {
20
+ "type": "VISUAL",
21
+ "shape": [
22
+ 3,
23
+ 480,
24
+ 640
25
+ ]
26
+ }
27
+ },
28
+ "output_features": {
29
+ "action": {
30
+ "type": "ACTION",
31
+ "shape": [
32
+ 6
33
+ ]
34
+ }
35
+ },
36
+ "device": "cuda",
37
+ "use_amp": false,
38
+ "use_peft": false,
39
+ "push_to_hub": false,
40
+ "repo_id": null,
41
+ "private": null,
42
+ "tags": null,
43
+ "license": null,
44
+ "pretrained_path": null,
45
+ "chunk_size": 90,
46
+ "n_action_steps": 30,
47
+ "normalization_mapping": {
48
+ "VISUAL": "MEAN_STD",
49
+ "STATE": "MEAN_STD",
50
+ "ACTION": "MEAN_STD"
51
+ },
52
+ "vision_backbone": "resnet18",
53
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
54
+ "replace_final_stride_with_dilation": false,
55
+ "pre_norm": false,
56
+ "dim_model": 512,
57
+ "n_heads": 8,
58
+ "dim_feedforward": 3200,
59
+ "feedforward_activation": "relu",
60
+ "n_encoder_layers": 4,
61
+ "n_decoder_layers": 1,
62
+ "use_vae": true,
63
+ "latent_dim": 32,
64
+ "n_vae_encoder_layers": 4,
65
+ "temporal_ensemble_coeff": null,
66
+ "dropout": 0.1,
67
+ "kl_weight": 10.0,
68
+ "optimizer_lr": 1e-05,
69
+ "optimizer_weight_decay": 0.0001,
70
+ "optimizer_lr_backbone": 1e-05
71
+ }
pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aafc224d3cc82ccb3715c362498c614a5d02d0a017b96a113623599729d2db87
3
+ size 206658768
pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/policy_postprocessor.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "unnormalizer_processor",
6
+ "config": {
7
+ "eps": 1e-08,
8
+ "features": {
9
+ "action": {
10
+ "type": "ACTION",
11
+ "shape": [
12
+ 6
13
+ ]
14
+ }
15
+ },
16
+ "norm_map": {
17
+ "VISUAL": "MEAN_STD",
18
+ "STATE": "MEAN_STD",
19
+ "ACTION": "MEAN_STD"
20
+ }
21
+ },
22
+ "state_file": "policy_postprocessor_step_0_unnormalizer_processor.safetensors"
23
+ },
24
+ {
25
+ "registry_name": "device_processor",
26
+ "config": {
27
+ "device": "cpu",
28
+ "float_dtype": null
29
+ }
30
+ }
31
+ ]
32
+ }
pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:931ed929c95ef00f6e334882511164d4c9d89f16bfb31eeb212ddecf99897ed7
3
+ size 7552
pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/policy_preprocessor.json ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {}
8
+ }
9
+ },
10
+ {
11
+ "registry_name": "to_batch_processor",
12
+ "config": {}
13
+ },
14
+ {
15
+ "registry_name": "device_processor",
16
+ "config": {
17
+ "device": "cuda",
18
+ "float_dtype": null
19
+ }
20
+ },
21
+ {
22
+ "registry_name": "normalizer_processor",
23
+ "config": {
24
+ "eps": 1e-08,
25
+ "features": {
26
+ "observation.state": {
27
+ "type": "STATE",
28
+ "shape": [
29
+ 6
30
+ ]
31
+ },
32
+ "observation.images.front": {
33
+ "type": "VISUAL",
34
+ "shape": [
35
+ 3,
36
+ 480,
37
+ 640
38
+ ]
39
+ },
40
+ "observation.images.wrist": {
41
+ "type": "VISUAL",
42
+ "shape": [
43
+ 3,
44
+ 480,
45
+ 640
46
+ ]
47
+ },
48
+ "action": {
49
+ "type": "ACTION",
50
+ "shape": [
51
+ 6
52
+ ]
53
+ }
54
+ },
55
+ "norm_map": {
56
+ "VISUAL": "MEAN_STD",
57
+ "STATE": "MEAN_STD",
58
+ "ACTION": "MEAN_STD"
59
+ }
60
+ },
61
+ "state_file": "policy_preprocessor_step_3_normalizer_processor.safetensors"
62
+ }
63
+ ]
64
+ }
pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:931ed929c95ef00f6e334882511164d4c9d89f16bfb31eeb212ddecf99897ed7
3
+ size 7552
pick_v3_1-2-pieces/checkpoints/024000/pretrained_model/train_config.json ADDED
@@ -0,0 +1,533 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "local/so101_act",
4
+ "root": "C:\\Users\\Usuario\\Documents\\Workspace\\LeRobot_\\demos\\06_imitation_learning\\data\\real_so101\\act_pick_super",
5
+ "episodes": [
6
+ 64,
7
+ 65,
8
+ 70,
9
+ 74,
10
+ 78,
11
+ 79,
12
+ 84,
13
+ 89,
14
+ 90,
15
+ 93,
16
+ 94,
17
+ 100,
18
+ 101,
19
+ 102,
20
+ 103,
21
+ 104,
22
+ 105,
23
+ 106,
24
+ 107,
25
+ 109,
26
+ 110,
27
+ 111,
28
+ 112,
29
+ 113,
30
+ 114,
31
+ 115,
32
+ 116,
33
+ 117,
34
+ 118,
35
+ 119,
36
+ 120,
37
+ 121,
38
+ 122,
39
+ 124,
40
+ 125,
41
+ 126,
42
+ 127,
43
+ 128,
44
+ 129,
45
+ 130,
46
+ 131,
47
+ 132,
48
+ 133,
49
+ 134,
50
+ 135,
51
+ 136,
52
+ 137,
53
+ 138,
54
+ 139,
55
+ 140,
56
+ 141,
57
+ 142,
58
+ 143,
59
+ 144,
60
+ 145,
61
+ 146,
62
+ 147,
63
+ 148,
64
+ 149,
65
+ 150,
66
+ 151,
67
+ 152,
68
+ 153,
69
+ 154,
70
+ 155,
71
+ 156,
72
+ 157,
73
+ 158,
74
+ 159,
75
+ 160,
76
+ 161,
77
+ 162,
78
+ 163,
79
+ 164,
80
+ 165,
81
+ 166,
82
+ 167,
83
+ 168,
84
+ 169,
85
+ 170,
86
+ 171,
87
+ 172,
88
+ 173,
89
+ 174,
90
+ 175,
91
+ 176,
92
+ 177,
93
+ 178,
94
+ 179,
95
+ 180,
96
+ 181,
97
+ 182,
98
+ 183,
99
+ 184,
100
+ 185,
101
+ 186,
102
+ 187,
103
+ 188,
104
+ 189,
105
+ 190,
106
+ 191,
107
+ 192,
108
+ 193,
109
+ 194,
110
+ 195,
111
+ 196,
112
+ 197,
113
+ 198,
114
+ 199,
115
+ 200,
116
+ 201,
117
+ 202,
118
+ 203,
119
+ 204,
120
+ 205,
121
+ 206,
122
+ 207,
123
+ 208,
124
+ 209,
125
+ 210,
126
+ 211,
127
+ 212,
128
+ 213,
129
+ 214,
130
+ 215,
131
+ 216,
132
+ 217,
133
+ 218,
134
+ 219,
135
+ 220,
136
+ 221,
137
+ 222,
138
+ 223,
139
+ 224,
140
+ 225,
141
+ 226,
142
+ 227,
143
+ 228,
144
+ 229,
145
+ 230,
146
+ 231,
147
+ 232,
148
+ 233,
149
+ 235,
150
+ 236,
151
+ 237,
152
+ 238,
153
+ 239,
154
+ 240,
155
+ 241,
156
+ 242,
157
+ 243,
158
+ 244,
159
+ 245,
160
+ 246,
161
+ 247,
162
+ 248,
163
+ 249,
164
+ 250,
165
+ 251,
166
+ 252,
167
+ 253,
168
+ 255,
169
+ 256,
170
+ 257,
171
+ 258,
172
+ 259,
173
+ 260,
174
+ 261,
175
+ 262,
176
+ 263,
177
+ 264,
178
+ 265,
179
+ 266,
180
+ 267,
181
+ 268,
182
+ 269,
183
+ 270,
184
+ 271,
185
+ 272,
186
+ 273,
187
+ 274,
188
+ 275,
189
+ 276,
190
+ 277,
191
+ 278,
192
+ 279,
193
+ 280,
194
+ 281,
195
+ 282,
196
+ 283,
197
+ 284,
198
+ 285,
199
+ 286,
200
+ 287,
201
+ 288,
202
+ 289,
203
+ 290,
204
+ 291,
205
+ 292,
206
+ 293,
207
+ 294,
208
+ 295,
209
+ 296,
210
+ 297,
211
+ 298,
212
+ 299,
213
+ 300,
214
+ 301,
215
+ 302,
216
+ 303,
217
+ 304,
218
+ 305,
219
+ 306,
220
+ 307,
221
+ 308,
222
+ 309,
223
+ 310,
224
+ 311,
225
+ 312,
226
+ 313,
227
+ 314,
228
+ 315,
229
+ 316,
230
+ 317,
231
+ 318,
232
+ 319,
233
+ 320,
234
+ 325,
235
+ 326,
236
+ 327,
237
+ 329,
238
+ 330,
239
+ 331,
240
+ 332,
241
+ 333,
242
+ 334,
243
+ 335,
244
+ 336,
245
+ 337,
246
+ 338,
247
+ 339,
248
+ 340,
249
+ 341,
250
+ 342,
251
+ 343,
252
+ 344,
253
+ 345,
254
+ 346,
255
+ 347,
256
+ 348,
257
+ 349,
258
+ 351,
259
+ 352,
260
+ 353,
261
+ 354,
262
+ 355,
263
+ 356,
264
+ 357,
265
+ 358,
266
+ 359,
267
+ 360,
268
+ 361,
269
+ 364,
270
+ 365,
271
+ 366,
272
+ 367,
273
+ 368,
274
+ 369,
275
+ 370,
276
+ 371,
277
+ 372,
278
+ 373,
279
+ 374,
280
+ 375,
281
+ 376,
282
+ 377,
283
+ 378,
284
+ 391,
285
+ 392,
286
+ 393,
287
+ 394,
288
+ 395,
289
+ 396,
290
+ 397,
291
+ 398,
292
+ 399,
293
+ 400,
294
+ 401,
295
+ 402,
296
+ 403,
297
+ 404,
298
+ 405,
299
+ 406,
300
+ 407,
301
+ 408,
302
+ 409,
303
+ 410,
304
+ 411,
305
+ 412,
306
+ 413,
307
+ 414,
308
+ 415,
309
+ 416,
310
+ 417,
311
+ 419,
312
+ 420,
313
+ 421,
314
+ 422,
315
+ 423,
316
+ 424,
317
+ 425,
318
+ 426,
319
+ 427,
320
+ 428,
321
+ 429,
322
+ 430,
323
+ 431,
324
+ 432,
325
+ 433,
326
+ 434,
327
+ 435,
328
+ 436,
329
+ 438,
330
+ 439,
331
+ 440,
332
+ 441,
333
+ 442,
334
+ 443,
335
+ 444,
336
+ 445
337
+ ],
338
+ "image_transforms": {
339
+ "enable": true,
340
+ "max_num_transforms": 3,
341
+ "random_order": false,
342
+ "tfs": {
343
+ "brightness": {
344
+ "weight": 1.0,
345
+ "type": "ColorJitter",
346
+ "kwargs": {
347
+ "brightness": [
348
+ 0.8,
349
+ 1.2
350
+ ]
351
+ }
352
+ },
353
+ "contrast": {
354
+ "weight": 1.0,
355
+ "type": "ColorJitter",
356
+ "kwargs": {
357
+ "contrast": [
358
+ 0.8,
359
+ 1.2
360
+ ]
361
+ }
362
+ },
363
+ "saturation": {
364
+ "weight": 1.0,
365
+ "type": "ColorJitter",
366
+ "kwargs": {
367
+ "saturation": [
368
+ 0.5,
369
+ 1.5
370
+ ]
371
+ }
372
+ },
373
+ "hue": {
374
+ "weight": 1.0,
375
+ "type": "ColorJitter",
376
+ "kwargs": {
377
+ "hue": [
378
+ -0.05,
379
+ 0.05
380
+ ]
381
+ }
382
+ },
383
+ "sharpness": {
384
+ "weight": 1.0,
385
+ "type": "SharpnessJitter",
386
+ "kwargs": {
387
+ "sharpness": [
388
+ 0.5,
389
+ 1.5
390
+ ]
391
+ }
392
+ },
393
+ "affine": {
394
+ "weight": 1.0,
395
+ "type": "RandomAffine",
396
+ "kwargs": {
397
+ "degrees": [
398
+ -5.0,
399
+ 5.0
400
+ ],
401
+ "translate": [
402
+ 0.05,
403
+ 0.05
404
+ ]
405
+ }
406
+ }
407
+ }
408
+ },
409
+ "revision": null,
410
+ "use_imagenet_stats": true,
411
+ "video_backend": "pyav",
412
+ "streaming": false
413
+ },
414
+ "env": null,
415
+ "policy": {
416
+ "type": "act",
417
+ "n_obs_steps": 1,
418
+ "input_features": {
419
+ "observation.state": {
420
+ "type": "STATE",
421
+ "shape": [
422
+ 6
423
+ ]
424
+ },
425
+ "observation.images.front": {
426
+ "type": "VISUAL",
427
+ "shape": [
428
+ 3,
429
+ 480,
430
+ 640
431
+ ]
432
+ },
433
+ "observation.images.wrist": {
434
+ "type": "VISUAL",
435
+ "shape": [
436
+ 3,
437
+ 480,
438
+ 640
439
+ ]
440
+ }
441
+ },
442
+ "output_features": {
443
+ "action": {
444
+ "type": "ACTION",
445
+ "shape": [
446
+ 6
447
+ ]
448
+ }
449
+ },
450
+ "device": "cuda",
451
+ "use_amp": false,
452
+ "use_peft": false,
453
+ "push_to_hub": false,
454
+ "repo_id": null,
455
+ "private": null,
456
+ "tags": null,
457
+ "license": null,
458
+ "pretrained_path": null,
459
+ "chunk_size": 90,
460
+ "n_action_steps": 30,
461
+ "normalization_mapping": {
462
+ "VISUAL": "MEAN_STD",
463
+ "STATE": "MEAN_STD",
464
+ "ACTION": "MEAN_STD"
465
+ },
466
+ "vision_backbone": "resnet18",
467
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
468
+ "replace_final_stride_with_dilation": false,
469
+ "pre_norm": false,
470
+ "dim_model": 512,
471
+ "n_heads": 8,
472
+ "dim_feedforward": 3200,
473
+ "feedforward_activation": "relu",
474
+ "n_encoder_layers": 4,
475
+ "n_decoder_layers": 1,
476
+ "use_vae": true,
477
+ "latent_dim": 32,
478
+ "n_vae_encoder_layers": 4,
479
+ "temporal_ensemble_coeff": null,
480
+ "dropout": 0.1,
481
+ "kl_weight": 10.0,
482
+ "optimizer_lr": 1e-05,
483
+ "optimizer_weight_decay": 0.0001,
484
+ "optimizer_lr_backbone": 1e-05
485
+ },
486
+ "output_dir": "C:\\Users\\Usuario\\Documents\\Workspace\\LeRobot_\\demos\\06_imitation_learning\\outputs\\sweep_v3",
487
+ "job_name": "act_so101_sweep_v3",
488
+ "resume": false,
489
+ "seed": 1000,
490
+ "num_workers": 4,
491
+ "batch_size": 8,
492
+ "steps": 30000,
493
+ "eval_freq": 20000,
494
+ "log_freq": 100,
495
+ "tolerance_s": 0.0001,
496
+ "save_checkpoint": true,
497
+ "save_freq": 6000,
498
+ "use_policy_training_preset": true,
499
+ "optimizer": {
500
+ "type": "adamw",
501
+ "lr": 1e-05,
502
+ "weight_decay": 0.0001,
503
+ "grad_clip_norm": 10.0,
504
+ "betas": [
505
+ 0.9,
506
+ 0.999
507
+ ],
508
+ "eps": 1e-08
509
+ },
510
+ "scheduler": null,
511
+ "eval": {
512
+ "n_episodes": 50,
513
+ "batch_size": 50,
514
+ "use_async_envs": false
515
+ },
516
+ "wandb": {
517
+ "enable": false,
518
+ "disable_artifact": false,
519
+ "project": "lerobot",
520
+ "entity": null,
521
+ "notes": null,
522
+ "run_id": null,
523
+ "mode": null
524
+ },
525
+ "peft": null,
526
+ "use_rabc": false,
527
+ "rabc_progress_path": null,
528
+ "rabc_kappa": 0.01,
529
+ "rabc_epsilon": 1e-06,
530
+ "rabc_head_mode": "sparse",
531
+ "rename_map": {},
532
+ "checkpoint_path": null
533
+ }
pick_v3_1-2-pieces/checkpoints/030000/dataset_features.json ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action": {
3
+ "dtype": "float32",
4
+ "names": [
5
+ "shoulder_pan.pos",
6
+ "shoulder_lift.pos",
7
+ "elbow_flex.pos",
8
+ "wrist_flex.pos",
9
+ "wrist_roll.pos",
10
+ "gripper.pos"
11
+ ],
12
+ "shape": [
13
+ 6
14
+ ]
15
+ },
16
+ "observation.images.front": {
17
+ "dtype": "video",
18
+ "shape": [
19
+ 480,
20
+ 640,
21
+ 3
22
+ ],
23
+ "names": [
24
+ "height",
25
+ "width",
26
+ "channels"
27
+ ],
28
+ "info": {
29
+ "is_depth_map": false,
30
+ "video.height": 480,
31
+ "video.width": 640,
32
+ "video.codec": "h264",
33
+ "video.pix_fmt": "yuv420p",
34
+ "video.fps": 30,
35
+ "video.channels": 3,
36
+ "has_audio": false,
37
+ "video.g": 2,
38
+ "video.crf": 30,
39
+ "video.preset": "p1",
40
+ "video.fast_decode": 0,
41
+ "video.video_backend": "pyav",
42
+ "video.extra_options": {}
43
+ }
44
+ },
45
+ "observation.images.wrist": {
46
+ "dtype": "video",
47
+ "shape": [
48
+ 480,
49
+ 640,
50
+ 3
51
+ ],
52
+ "names": [
53
+ "height",
54
+ "width",
55
+ "channels"
56
+ ],
57
+ "info": {
58
+ "is_depth_map": false,
59
+ "video.height": 480,
60
+ "video.width": 640,
61
+ "video.codec": "h264",
62
+ "video.pix_fmt": "yuv420p",
63
+ "video.fps": 30,
64
+ "video.channels": 3,
65
+ "has_audio": false,
66
+ "video.g": 2,
67
+ "video.crf": 30,
68
+ "video.preset": "p1",
69
+ "video.fast_decode": 0,
70
+ "video.video_backend": "pyav",
71
+ "video.extra_options": {}
72
+ }
73
+ },
74
+ "observation.state": {
75
+ "dtype": "float32",
76
+ "names": [
77
+ "shoulder_pan.pos",
78
+ "shoulder_lift.pos",
79
+ "elbow_flex.pos",
80
+ "wrist_flex.pos",
81
+ "wrist_roll.pos",
82
+ "gripper.pos"
83
+ ],
84
+ "shape": [
85
+ 6
86
+ ]
87
+ }
88
+ }
pick_v3_1-2-pieces/checkpoints/030000/model_card_original.md ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: lerobot
4
+ pipeline_tag: robotics
5
+ tags:
6
+ - lerobot
7
+ - act
8
+ - so101
9
+ - pick
10
+ ---
11
+
12
+ # act-pick-v3-1-2-pieces-30000
13
+
14
+ The v2 recipe plus the two new evening sessions of 24 Oct (third round, v3), 30,000-step run. Checkpoint at 30,000 steps. This is not the best checkpoint of its run (the best one is at 18,000 steps); it is published for completeness.
15
+
16
+ ACT policy for the SO-101 "pick a piece from the crate and hold it up to the camera" task. Inputs: front camera, wrist camera and the six joint positions; output: a chunk of 90 future joint targets at 30 fps.
17
+
18
+ ## Training data
19
+
20
+ 331 training episodes, 30 fps only, one or two pieces in the crate at the start (every 24 Oct session, plus 12 episodes of 23 Oct), 5 to 15 s long, start gap at most 20 degrees, at most 3 s idle at either end, gripper opens at least once, known bad takes removed.
21
+
22
+ Recipe: LeRobot `train()`, batch 8, chunk 90, 30 action steps per replan, default image augmentation, seed 1000.
23
+
24
+ ## Offline check on the held-out episodes
25
+
26
+ Held out: 16 episodes, 12 from `so101-pick-cube_20261024_181939` (one piece) and 4 from `so101-pick-cube_20261024_144537` (two pieces). Mean absolute joint error over the predicted chunk (lower is better) against a "stay at the current joint positions" baseline of 20.71:
27
+
28
+ | checkpoint of this run | error | below baseline |
29
+ |---|---|---|
30
+ | 6,000 steps | 15.86 | 23% |
31
+ | 12,000 steps | 15.89 | 23% |
32
+ | 18,000 steps | 15.35 | 26% |
33
+ | 24,000 steps | 15.73 | 24% |
34
+ | **30,000 steps (this one)** | **16.38** | 21% |
35
+
36
+ All runs of this round, best checkpoint of each:
37
+
38
+ | run | training episodes | best checkpoint | error | below stay-still | repo |
39
+ |---|---|---|---|---|---|
40
+ | super (all quality-passing data) | 398 | 6k | 15.16 | 27% | act-pick-v3-6000 |
41
+ | strict (6-12 s, tight quality) | 205 | 2.4k | 15.14 | 27% | act-pick-v3-strict-2400 |
42
+ | v3 recipe on 1-2 piece episodes | 331 | 18k | 15.35 | 26% | act-pick-v3-1-2-pieces-18000 |
43
+ | v2 (earlier) | 278 | 6k | 15.36 | 26% | act-pick-v2-6000 |
44
+ | chunk 60 (own baseline 16.23) | 331 | 4.8k | 11.94 | 26% | act-pick-v3-chunk60-4800 |
45
+ | one piece only | 292 | 9.6k | 15.47 | 25% | act-pick-v3-1piece-9600 |
46
+ | recent sessions only | 100 | 12k | 16.87 | 19% | act-pick-v3-recent-12000 |
47
+ | first run | 165 | 6k | 16.87 | 19% | act-pick-* |
48
+
49
+ The best checkpoints of the main runs are within noise of each other, and every run got worse after about one pass over the data. The offline error cannot say which model grasps best; none of them has been tested on the robot.
pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/config.json ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "act",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.state": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 6
9
+ ]
10
+ },
11
+ "observation.images.front": {
12
+ "type": "VISUAL",
13
+ "shape": [
14
+ 3,
15
+ 480,
16
+ 640
17
+ ]
18
+ },
19
+ "observation.images.wrist": {
20
+ "type": "VISUAL",
21
+ "shape": [
22
+ 3,
23
+ 480,
24
+ 640
25
+ ]
26
+ }
27
+ },
28
+ "output_features": {
29
+ "action": {
30
+ "type": "ACTION",
31
+ "shape": [
32
+ 6
33
+ ]
34
+ }
35
+ },
36
+ "device": "cuda",
37
+ "use_amp": false,
38
+ "use_peft": false,
39
+ "push_to_hub": false,
40
+ "repo_id": null,
41
+ "private": null,
42
+ "tags": null,
43
+ "license": null,
44
+ "pretrained_path": null,
45
+ "chunk_size": 90,
46
+ "n_action_steps": 30,
47
+ "normalization_mapping": {
48
+ "VISUAL": "MEAN_STD",
49
+ "STATE": "MEAN_STD",
50
+ "ACTION": "MEAN_STD"
51
+ },
52
+ "vision_backbone": "resnet18",
53
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
54
+ "replace_final_stride_with_dilation": false,
55
+ "pre_norm": false,
56
+ "dim_model": 512,
57
+ "n_heads": 8,
58
+ "dim_feedforward": 3200,
59
+ "feedforward_activation": "relu",
60
+ "n_encoder_layers": 4,
61
+ "n_decoder_layers": 1,
62
+ "use_vae": true,
63
+ "latent_dim": 32,
64
+ "n_vae_encoder_layers": 4,
65
+ "temporal_ensemble_coeff": null,
66
+ "dropout": 0.1,
67
+ "kl_weight": 10.0,
68
+ "optimizer_lr": 1e-05,
69
+ "optimizer_weight_decay": 0.0001,
70
+ "optimizer_lr_backbone": 1e-05
71
+ }
pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2a44a101d2ccaecad998c67706e076f2a6186b03b968ce358f50d65ad49e1715
3
+ size 206658768
pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/policy_postprocessor.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "unnormalizer_processor",
6
+ "config": {
7
+ "eps": 1e-08,
8
+ "features": {
9
+ "action": {
10
+ "type": "ACTION",
11
+ "shape": [
12
+ 6
13
+ ]
14
+ }
15
+ },
16
+ "norm_map": {
17
+ "VISUAL": "MEAN_STD",
18
+ "STATE": "MEAN_STD",
19
+ "ACTION": "MEAN_STD"
20
+ }
21
+ },
22
+ "state_file": "policy_postprocessor_step_0_unnormalizer_processor.safetensors"
23
+ },
24
+ {
25
+ "registry_name": "device_processor",
26
+ "config": {
27
+ "device": "cpu",
28
+ "float_dtype": null
29
+ }
30
+ }
31
+ ]
32
+ }
pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:931ed929c95ef00f6e334882511164d4c9d89f16bfb31eeb212ddecf99897ed7
3
+ size 7552
pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/policy_preprocessor.json ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {}
8
+ }
9
+ },
10
+ {
11
+ "registry_name": "to_batch_processor",
12
+ "config": {}
13
+ },
14
+ {
15
+ "registry_name": "device_processor",
16
+ "config": {
17
+ "device": "cuda",
18
+ "float_dtype": null
19
+ }
20
+ },
21
+ {
22
+ "registry_name": "normalizer_processor",
23
+ "config": {
24
+ "eps": 1e-08,
25
+ "features": {
26
+ "observation.state": {
27
+ "type": "STATE",
28
+ "shape": [
29
+ 6
30
+ ]
31
+ },
32
+ "observation.images.front": {
33
+ "type": "VISUAL",
34
+ "shape": [
35
+ 3,
36
+ 480,
37
+ 640
38
+ ]
39
+ },
40
+ "observation.images.wrist": {
41
+ "type": "VISUAL",
42
+ "shape": [
43
+ 3,
44
+ 480,
45
+ 640
46
+ ]
47
+ },
48
+ "action": {
49
+ "type": "ACTION",
50
+ "shape": [
51
+ 6
52
+ ]
53
+ }
54
+ },
55
+ "norm_map": {
56
+ "VISUAL": "MEAN_STD",
57
+ "STATE": "MEAN_STD",
58
+ "ACTION": "MEAN_STD"
59
+ }
60
+ },
61
+ "state_file": "policy_preprocessor_step_3_normalizer_processor.safetensors"
62
+ }
63
+ ]
64
+ }
pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/policy_preprocessor_step_3_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:931ed929c95ef00f6e334882511164d4c9d89f16bfb31eeb212ddecf99897ed7
3
+ size 7552
pick_v3_1-2-pieces/checkpoints/030000/pretrained_model/train_config.json ADDED
@@ -0,0 +1,533 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "local/so101_act",
4
+ "root": "C:\\Users\\Usuario\\Documents\\Workspace\\LeRobot_\\demos\\06_imitation_learning\\data\\real_so101\\act_pick_super",
5
+ "episodes": [
6
+ 64,
7
+ 65,
8
+ 70,
9
+ 74,
10
+ 78,
11
+ 79,
12
+ 84,
13
+ 89,
14
+ 90,
15
+ 93,
16
+ 94,
17
+ 100,
18
+ 101,
19
+ 102,
20
+ 103,
21
+ 104,
22
+ 105,
23
+ 106,
24
+ 107,
25
+ 109,
26
+ 110,
27
+ 111,
28
+ 112,
29
+ 113,
30
+ 114,
31
+ 115,
32
+ 116,
33
+ 117,
34
+ 118,
35
+ 119,
36
+ 120,
37
+ 121,
38
+ 122,
39
+ 124,
40
+ 125,
41
+ 126,
42
+ 127,
43
+ 128,
44
+ 129,
45
+ 130,
46
+ 131,
47
+ 132,
48
+ 133,
49
+ 134,
50
+ 135,
51
+ 136,
52
+ 137,
53
+ 138,
54
+ 139,
55
+ 140,
56
+ 141,
57
+ 142,
58
+ 143,
59
+ 144,
60
+ 145,
61
+ 146,
62
+ 147,
63
+ 148,
64
+ 149,
65
+ 150,
66
+ 151,
67
+ 152,
68
+ 153,
69
+ 154,
70
+ 155,
71
+ 156,
72
+ 157,
73
+ 158,
74
+ 159,
75
+ 160,
76
+ 161,
77
+ 162,
78
+ 163,
79
+ 164,
80
+ 165,
81
+ 166,
82
+ 167,
83
+ 168,
84
+ 169,
85
+ 170,
86
+ 171,
87
+ 172,
88
+ 173,
89
+ 174,
90
+ 175,
91
+ 176,
92
+ 177,
93
+ 178,
94
+ 179,
95
+ 180,
96
+ 181,
97
+ 182,
98
+ 183,
99
+ 184,
100
+ 185,
101
+ 186,
102
+ 187,
103
+ 188,
104
+ 189,
105
+ 190,
106
+ 191,
107
+ 192,
108
+ 193,
109
+ 194,
110
+ 195,
111
+ 196,
112
+ 197,
113
+ 198,
114
+ 199,
115
+ 200,
116
+ 201,
117
+ 202,
118
+ 203,
119
+ 204,
120
+ 205,
121
+ 206,
122
+ 207,
123
+ 208,
124
+ 209,
125
+ 210,
126
+ 211,
127
+ 212,
128
+ 213,
129
+ 214,
130
+ 215,
131
+ 216,
132
+ 217,
133
+ 218,
134
+ 219,
135
+ 220,
136
+ 221,
137
+ 222,
138
+ 223,
139
+ 224,
140
+ 225,
141
+ 226,
142
+ 227,
143
+ 228,
144
+ 229,
145
+ 230,
146
+ 231,
147
+ 232,
148
+ 233,
149
+ 235,
150
+ 236,
151
+ 237,
152
+ 238,
153
+ 239,
154
+ 240,
155
+ 241,
156
+ 242,
157
+ 243,
158
+ 244,
159
+ 245,
160
+ 246,
161
+ 247,
162
+ 248,
163
+ 249,
164
+ 250,
165
+ 251,
166
+ 252,
167
+ 253,
168
+ 255,
169
+ 256,
170
+ 257,
171
+ 258,
172
+ 259,
173
+ 260,
174
+ 261,
175
+ 262,
176
+ 263,
177
+ 264,
178
+ 265,
179
+ 266,
180
+ 267,
181
+ 268,
182
+ 269,
183
+ 270,
184
+ 271,
185
+ 272,
186
+ 273,
187
+ 274,
188
+ 275,
189
+ 276,
190
+ 277,
191
+ 278,
192
+ 279,
193
+ 280,
194
+ 281,
195
+ 282,
196
+ 283,
197
+ 284,
198
+ 285,
199
+ 286,
200
+ 287,
201
+ 288,
202
+ 289,
203
+ 290,
204
+ 291,
205
+ 292,
206
+ 293,
207
+ 294,
208
+ 295,
209
+ 296,
210
+ 297,
211
+ 298,
212
+ 299,
213
+ 300,
214
+ 301,
215
+ 302,
216
+ 303,
217
+ 304,
218
+ 305,
219
+ 306,
220
+ 307,
221
+ 308,
222
+ 309,
223
+ 310,
224
+ 311,
225
+ 312,
226
+ 313,
227
+ 314,
228
+ 315,
229
+ 316,
230
+ 317,
231
+ 318,
232
+ 319,
233
+ 320,
234
+ 325,
235
+ 326,
236
+ 327,
237
+ 329,
238
+ 330,
239
+ 331,
240
+ 332,
241
+ 333,
242
+ 334,
243
+ 335,
244
+ 336,
245
+ 337,
246
+ 338,
247
+ 339,
248
+ 340,
249
+ 341,
250
+ 342,
251
+ 343,
252
+ 344,
253
+ 345,
254
+ 346,
255
+ 347,
256
+ 348,
257
+ 349,
258
+ 351,
259
+ 352,
260
+ 353,
261
+ 354,
262
+ 355,
263
+ 356,
264
+ 357,
265
+ 358,
266
+ 359,
267
+ 360,
268
+ 361,
269
+ 364,
270
+ 365,
271
+ 366,
272
+ 367,
273
+ 368,
274
+ 369,
275
+ 370,
276
+ 371,
277
+ 372,
278
+ 373,
279
+ 374,
280
+ 375,
281
+ 376,
282
+ 377,
283
+ 378,
284
+ 391,
285
+ 392,
286
+ 393,
287
+ 394,
288
+ 395,
289
+ 396,
290
+ 397,
291
+ 398,
292
+ 399,
293
+ 400,
294
+ 401,
295
+ 402,
296
+ 403,
297
+ 404,
298
+ 405,
299
+ 406,
300
+ 407,
301
+ 408,
302
+ 409,
303
+ 410,
304
+ 411,
305
+ 412,
306
+ 413,
307
+ 414,
308
+ 415,
309
+ 416,
310
+ 417,
311
+ 419,
312
+ 420,
313
+ 421,
314
+ 422,
315
+ 423,
316
+ 424,
317
+ 425,
318
+ 426,
319
+ 427,
320
+ 428,
321
+ 429,
322
+ 430,
323
+ 431,
324
+ 432,
325
+ 433,
326
+ 434,
327
+ 435,
328
+ 436,
329
+ 438,
330
+ 439,
331
+ 440,
332
+ 441,
333
+ 442,
334
+ 443,
335
+ 444,
336
+ 445
337
+ ],
338
+ "image_transforms": {
339
+ "enable": true,
340
+ "max_num_transforms": 3,
341
+ "random_order": false,
342
+ "tfs": {
343
+ "brightness": {
344
+ "weight": 1.0,
345
+ "type": "ColorJitter",
346
+ "kwargs": {
347
+ "brightness": [
348
+ 0.8,
349
+ 1.2
350
+ ]
351
+ }
352
+ },
353
+ "contrast": {
354
+ "weight": 1.0,
355
+ "type": "ColorJitter",
356
+ "kwargs": {
357
+ "contrast": [
358
+ 0.8,
359
+ 1.2
360
+ ]
361
+ }
362
+ },
363
+ "saturation": {
364
+ "weight": 1.0,
365
+ "type": "ColorJitter",
366
+ "kwargs": {
367
+ "saturation": [
368
+ 0.5,
369
+ 1.5
370
+ ]
371
+ }
372
+ },
373
+ "hue": {
374
+ "weight": 1.0,
375
+ "type": "ColorJitter",
376
+ "kwargs": {
377
+ "hue": [
378
+ -0.05,
379
+ 0.05
380
+ ]
381
+ }
382
+ },
383
+ "sharpness": {
384
+ "weight": 1.0,
385
+ "type": "SharpnessJitter",
386
+ "kwargs": {
387
+ "sharpness": [
388
+ 0.5,
389
+ 1.5
390
+ ]
391
+ }
392
+ },
393
+ "affine": {
394
+ "weight": 1.0,
395
+ "type": "RandomAffine",
396
+ "kwargs": {
397
+ "degrees": [
398
+ -5.0,
399
+ 5.0
400
+ ],
401
+ "translate": [
402
+ 0.05,
403
+ 0.05
404
+ ]
405
+ }
406
+ }
407
+ }
408
+ },
409
+ "revision": null,
410
+ "use_imagenet_stats": true,
411
+ "video_backend": "pyav",
412
+ "streaming": false
413
+ },
414
+ "env": null,
415
+ "policy": {
416
+ "type": "act",
417
+ "n_obs_steps": 1,
418
+ "input_features": {
419
+ "observation.state": {
420
+ "type": "STATE",
421
+ "shape": [
422
+ 6
423
+ ]
424
+ },
425
+ "observation.images.front": {
426
+ "type": "VISUAL",
427
+ "shape": [
428
+ 3,
429
+ 480,
430
+ 640
431
+ ]
432
+ },
433
+ "observation.images.wrist": {
434
+ "type": "VISUAL",
435
+ "shape": [
436
+ 3,
437
+ 480,
438
+ 640
439
+ ]
440
+ }
441
+ },
442
+ "output_features": {
443
+ "action": {
444
+ "type": "ACTION",
445
+ "shape": [
446
+ 6
447
+ ]
448
+ }
449
+ },
450
+ "device": "cuda",
451
+ "use_amp": false,
452
+ "use_peft": false,
453
+ "push_to_hub": false,
454
+ "repo_id": null,
455
+ "private": null,
456
+ "tags": null,
457
+ "license": null,
458
+ "pretrained_path": null,
459
+ "chunk_size": 90,
460
+ "n_action_steps": 30,
461
+ "normalization_mapping": {
462
+ "VISUAL": "MEAN_STD",
463
+ "STATE": "MEAN_STD",
464
+ "ACTION": "MEAN_STD"
465
+ },
466
+ "vision_backbone": "resnet18",
467
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
468
+ "replace_final_stride_with_dilation": false,
469
+ "pre_norm": false,
470
+ "dim_model": 512,
471
+ "n_heads": 8,
472
+ "dim_feedforward": 3200,
473
+ "feedforward_activation": "relu",
474
+ "n_encoder_layers": 4,
475
+ "n_decoder_layers": 1,
476
+ "use_vae": true,
477
+ "latent_dim": 32,
478
+ "n_vae_encoder_layers": 4,
479
+ "temporal_ensemble_coeff": null,
480
+ "dropout": 0.1,
481
+ "kl_weight": 10.0,
482
+ "optimizer_lr": 1e-05,
483
+ "optimizer_weight_decay": 0.0001,
484
+ "optimizer_lr_backbone": 1e-05
485
+ },
486
+ "output_dir": "C:\\Users\\Usuario\\Documents\\Workspace\\LeRobot_\\demos\\06_imitation_learning\\outputs\\sweep_v3",
487
+ "job_name": "act_so101_sweep_v3",
488
+ "resume": false,
489
+ "seed": 1000,
490
+ "num_workers": 4,
491
+ "batch_size": 8,
492
+ "steps": 30000,
493
+ "eval_freq": 20000,
494
+ "log_freq": 100,
495
+ "tolerance_s": 0.0001,
496
+ "save_checkpoint": true,
497
+ "save_freq": 6000,
498
+ "use_policy_training_preset": true,
499
+ "optimizer": {
500
+ "type": "adamw",
501
+ "lr": 1e-05,
502
+ "weight_decay": 0.0001,
503
+ "grad_clip_norm": 10.0,
504
+ "betas": [
505
+ 0.9,
506
+ 0.999
507
+ ],
508
+ "eps": 1e-08
509
+ },
510
+ "scheduler": null,
511
+ "eval": {
512
+ "n_episodes": 50,
513
+ "batch_size": 50,
514
+ "use_async_envs": false
515
+ },
516
+ "wandb": {
517
+ "enable": false,
518
+ "disable_artifact": false,
519
+ "project": "lerobot",
520
+ "entity": null,
521
+ "notes": null,
522
+ "run_id": null,
523
+ "mode": null
524
+ },
525
+ "peft": null,
526
+ "use_rabc": false,
527
+ "rabc_progress_path": null,
528
+ "rabc_kappa": 0.01,
529
+ "rabc_epsilon": 1e-06,
530
+ "rabc_head_mode": "sparse",
531
+ "rename_map": {},
532
+ "checkpoint_path": null
533
+ }