Hugging Face's logo

deformable-bench
/
act-flatten-tshirt

Robotics
LeRobot
Safetensors
act
bimanual
deformable-object-manipulation
cloth
Model card Files
xet
 Community
hwk0809 commited on
Commit
e7312e4
·
verified ·
1 Parent(s): 377c8fd

ACT policy for flatten_tshirt (30k steps, 200-episode dataset)

Browse files
Files changed (8) hide show
  1. README.md +70 -0
  2. config.json +78 -0
  3. model.safetensors +3 -0
  4. policy_postprocessor.json +32 -0
  5. policy_postprocessor_step_0_unnormalizer_processor.safetensors +3 -0
  6. policy_preprocessor.json +72 -0
  7. policy_preprocessor_step_3_normalizer_processor.safetensors +3 -0
  8. train_config.json +195 -0
README.md ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: lerobot
4
+ pipeline_tag: robotics
5
+ tags:
6
+ - robotics
7
+ - lerobot
8
+ - act
9
+ - bimanual
10
+ - deformable-object-manipulation
11
+ - cloth
12
+ ---
13
+
14
+ # ACT — flatten_tshirt (bimanual cloth flattening)
15
+
16
+ An [ACT](https://arxiv.org/abs/2304.13705) policy trained on the `flatten_tshirt` task of a
17
+ bimanual deformable-object (cloth / bag) manipulation benchmark. The robot is a dual-arm
18
+ Piper; the task is to flatten a crumpled t-shirt on a table.
19
+
20
+ Simulation uses a GPU cloth solver co-simulated with the robot in a single model, and
21
+ observations are rendered with a photorealistic renderer.
22
+
23
+ ## Model
24
+
25
+ | | |
26
+ |---|---|
27
+ | Architecture | ACT, ResNet-18 vision backbone |
28
+ | Observation | 3 × RGB `720×1280` (`static_cam`, `left_hand_cam`, `right_hand_cam`) + 14-D joint state |
29
+ | Action | 14-D (left 6 joints + gripper, right 6 joints + gripper) |
30
+ | Chunk size / action steps | 100 / 100 |
31
+ | `n_obs_steps` | 1 |
32
+
33
+ ## Training
34
+
35
+ | | |
36
+ |---|---|
37
+ | Dataset | `flatten_tshirt_200` — 200 episodes / 41,464 frames, LeRobot v3.0, 25 fps |
38
+ | Steps | 30,000 |
39
+ | Batch size | 16 (single A100-80G) |
40
+ | Learning rate | 1e-5 |
41
+ | Seed | 1000 |
42
+ | Image augmentation | enabled, max 3 random transforms per sample |
43
+
44
+ Augmentation follows a tuned recipe (brightness / contrast / saturation / hue / sharpness /
45
+ small affine) rather than the LeRobot default, which is disabled. The brightness range is
46
+ deliberately asymmetric toward the darker side: simulation lighting is idealized while real
47
+ RealSense D435i footage tends to be darker, so biasing the augmentation toward darker samples
48
+ is the right direction for sim-to-real transfer.
49
+
50
+ ## Status
51
+
52
+ ⚠️ **This checkpoint has not yet been formally evaluated.** It has only been through a
53
+ small smoke-level closed-loop run, not the benchmark's standard N=100 protocol. Success-rate
54
+ numbers are deliberately not published here yet; they will be added once the full evaluation
55
+ has been run. Treat this as a training artifact, not a reported result.
56
+
57
+ ## Usage
58
+
59
+ ```python
60
+ from lerobot.policies.act.modeling_act import ACTPolicy
61
+
62
+ policy = ACTPolicy.from_pretrained("hwk0809/act-flatten-tshirt")
63
+ ```
64
+
65
+ The policy expects the three camera streams named exactly as listed above, plus a 14-D
66
+ `observation.state`, and returns a 14-D action. It runs in-process (no policy server needed).
67
+
68
+ ## License
69
+
70
+ Apache-2.0.
config.json ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "act",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.state": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 14
9
+ ]
10
+ },
11
+ "observation.images.static_cam": {
12
+ "type": "VISUAL",
13
+ "shape": [
14
+ 3,
15
+ 720,
16
+ 1280
17
+ ]
18
+ },
19
+ "observation.images.right_hand_cam": {
20
+ "type": "VISUAL",
21
+ "shape": [
22
+ 3,
23
+ 720,
24
+ 1280
25
+ ]
26
+ },
27
+ "observation.images.left_hand_cam": {
28
+ "type": "VISUAL",
29
+ "shape": [
30
+ 3,
31
+ 720,
32
+ 1280
33
+ ]
34
+ }
35
+ },
36
+ "output_features": {
37
+ "action": {
38
+ "type": "ACTION",
39
+ "shape": [
40
+ 14
41
+ ]
42
+ }
43
+ },
44
+ "device": "cuda",
45
+ "use_amp": false,
46
+ "push_to_hub": false,
47
+ "repo_id": null,
48
+ "private": null,
49
+ "tags": null,
50
+ "license": null,
51
+ "pretrained_path": null,
52
+ "chunk_size": 100,
53
+ "n_action_steps": 100,
54
+ "normalization_mapping": {
55
+ "VISUAL": "MEAN_STD",
56
+ "STATE": "MEAN_STD",
57
+ "ACTION": "MEAN_STD"
58
+ },
59
+ "vision_backbone": "resnet18",
60
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
61
+ "replace_final_stride_with_dilation": false,
62
+ "pre_norm": false,
63
+ "dim_model": 512,
64
+ "n_heads": 8,
65
+ "dim_feedforward": 3200,
66
+ "feedforward_activation": "relu",
67
+ "n_encoder_layers": 4,
68
+ "n_decoder_layers": 1,
69
+ "use_vae": true,
70
+ "latent_dim": 32,
71
+ "n_vae_encoder_layers": 4,
72
+ "temporal_ensemble_coeff": null,
73
+ "dropout": 0.1,
74
+ "kl_weight": 10.0,
75
+ "optimizer_lr": 1e-05,
76
+ "optimizer_weight_decay": 0.0001,
77
+ "optimizer_lr_backbone": 1e-05
78
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4640b3d8c3fc3b1899da444ed29dcc7b34d3dc5710d099470d760634f5a5f839
3
+ size 206765304
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
+ 14
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
+ }
policy_postprocessor_step_0_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ba3acc3f5c05bcc4b137e6cd973775f8c66cb8b8c6d735ada078df38b756d11
3
+ size 9416
policy_preprocessor.json ADDED
@@ -0,0 +1,72 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ 14
30
+ ]
31
+ },
32
+ "observation.images.static_cam": {
33
+ "type": "VISUAL",
34
+ "shape": [
35
+ 3,
36
+ 720,
37
+ 1280
38
+ ]
39
+ },
40
+ "observation.images.right_hand_cam": {
41
+ "type": "VISUAL",
42
+ "shape": [
43
+ 3,
44
+ 720,
45
+ 1280
46
+ ]
47
+ },
48
+ "observation.images.left_hand_cam": {
49
+ "type": "VISUAL",
50
+ "shape": [
51
+ 3,
52
+ 720,
53
+ 1280
54
+ ]
55
+ },
56
+ "action": {
57
+ "type": "ACTION",
58
+ "shape": [
59
+ 14
60
+ ]
61
+ }
62
+ },
63
+ "norm_map": {
64
+ "VISUAL": "MEAN_STD",
65
+ "STATE": "MEAN_STD",
66
+ "ACTION": "MEAN_STD"
67
+ }
68
+ },
69
+ "state_file": "policy_preprocessor_step_3_normalizer_processor.safetensors"
70
+ }
71
+ ]
72
+ }
policy_preprocessor_step_3_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ba3acc3f5c05bcc4b137e6cd973775f8c66cb8b8c6d735ada078df38b756d11
3
+ size 9416
train_config.json ADDED
@@ -0,0 +1,195 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "flatten_tshirt_200",
4
+ "root": null,
5
+ "episodes": null,
6
+ "image_transforms": {
7
+ "enable": true,
8
+ "max_num_transforms": 3,
9
+ "random_order": true,
10
+ "tfs": {
11
+ "brightness": {
12
+ "weight": 3.0,
13
+ "type": "ColorJitter",
14
+ "kwargs": {
15
+ "brightness": [
16
+ 0.75,
17
+ 1.15
18
+ ]
19
+ }
20
+ },
21
+ "contrast": {
22
+ "weight": 1.0,
23
+ "type": "ColorJitter",
24
+ "kwargs": {
25
+ "contrast": [
26
+ 0.8,
27
+ 1.2
28
+ ]
29
+ }
30
+ },
31
+ "saturation": {
32
+ "weight": 1.0,
33
+ "type": "ColorJitter",
34
+ "kwargs": {
35
+ "saturation": [
36
+ 0.5,
37
+ 1.5
38
+ ]
39
+ }
40
+ },
41
+ "hue": {
42
+ "weight": 3.0,
43
+ "type": "ColorJitter",
44
+ "kwargs": {
45
+ "hue": [
46
+ -0.15,
47
+ 0.15
48
+ ]
49
+ }
50
+ },
51
+ "sharpness": {
52
+ "weight": 1.0,
53
+ "type": "SharpnessJitter",
54
+ "kwargs": {
55
+ "sharpness": [
56
+ 0.5,
57
+ 1.5
58
+ ]
59
+ }
60
+ },
61
+ "affine": {
62
+ "weight": 1.0,
63
+ "type": "RandomAffine",
64
+ "kwargs": {
65
+ "degrees": [
66
+ -5.0,
67
+ 5.0
68
+ ],
69
+ "translate": [
70
+ 0.05,
71
+ 0.05
72
+ ]
73
+ }
74
+ }
75
+ }
76
+ },
77
+ "revision": null,
78
+ "use_imagenet_stats": true,
79
+ "video_backend": "torchcodec",
80
+ "streaming": false
81
+ },
82
+ "env": null,
83
+ "policy": {
84
+ "type": "act",
85
+ "n_obs_steps": 1,
86
+ "input_features": {
87
+ "observation.state": {
88
+ "type": "STATE",
89
+ "shape": [
90
+ 14
91
+ ]
92
+ },
93
+ "observation.images.static_cam": {
94
+ "type": "VISUAL",
95
+ "shape": [
96
+ 3,
97
+ 720,
98
+ 1280
99
+ ]
100
+ },
101
+ "observation.images.right_hand_cam": {
102
+ "type": "VISUAL",
103
+ "shape": [
104
+ 3,
105
+ 720,
106
+ 1280
107
+ ]
108
+ },
109
+ "observation.images.left_hand_cam": {
110
+ "type": "VISUAL",
111
+ "shape": [
112
+ 3,
113
+ 720,
114
+ 1280
115
+ ]
116
+ }
117
+ },
118
+ "output_features": {
119
+ "action": {
120
+ "type": "ACTION",
121
+ "shape": [
122
+ 14
123
+ ]
124
+ }
125
+ },
126
+ "device": "cuda",
127
+ "use_amp": false,
128
+ "push_to_hub": false,
129
+ "repo_id": null,
130
+ "private": null,
131
+ "tags": null,
132
+ "license": null,
133
+ "pretrained_path": null,
134
+ "chunk_size": 100,
135
+ "n_action_steps": 100,
136
+ "normalization_mapping": {
137
+ "VISUAL": "MEAN_STD",
138
+ "STATE": "MEAN_STD",
139
+ "ACTION": "MEAN_STD"
140
+ },
141
+ "vision_backbone": "resnet18",
142
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
143
+ "replace_final_stride_with_dilation": false,
144
+ "pre_norm": false,
145
+ "dim_model": 512,
146
+ "n_heads": 8,
147
+ "dim_feedforward": 3200,
148
+ "feedforward_activation": "relu",
149
+ "n_encoder_layers": 4,
150
+ "n_decoder_layers": 1,
151
+ "use_vae": true,
152
+ "latent_dim": 32,
153
+ "n_vae_encoder_layers": 4,
154
+ "temporal_ensemble_coeff": null,
155
+ "dropout": 0.1,
156
+ "kl_weight": 10.0,
157
+ "optimizer_lr": 1e-05,
158
+ "optimizer_weight_decay": 0.0001,
159
+ "optimizer_lr_backbone": 1e-05
160
+ },
161
+ "output_dir": null,
162
+ "job_name": "act_ft",
163
+ "resume": false,
164
+ "seed": 1000,
165
+ "num_workers": 8,
166
+ "batch_size": 16,
167
+ "steps": 30000,
168
+ "eval_freq": 20000,
169
+ "log_freq": 50,
170
+ "save_checkpoint": true,
171
+ "save_freq": 5000,
172
+ "use_policy_training_preset": true,
173
+ "optimizer": {
174
+ "type": "adamw",
175
+ "lr": 1e-05,
176
+ "weight_decay": 0.0001,
177
+ "grad_clip_norm": 10.0,
178
+ "betas": [
179
+ 0.9,
180
+ 0.999
181
+ ],
182
+ "eps": 1e-08
183
+ },
184
+ "scheduler": null,
185
+ "eval": {
186
+ "n_episodes": 50,
187
+ "batch_size": 50,
188
+ "use_async_envs": false
189
+ },
190
+ "wandb": {
191
+ "enable": false
192
+ },
193
+ "checkpoint_path": null,
194
+ "rename_map": {}
195
+ }