Robotics
LeRobot
Safetensors
diffusion
inaas commited on
Commit
93b45a6
·
verified ·
1 Parent(s): 33412c6

Upload policy weights, train config and readme

Browse files
Files changed (4) hide show
  1. README.md +62 -0
  2. config.json +95 -0
  3. model.safetensors +3 -0
  4. train_config.json +229 -0
README.md ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ datasets: inaas/push_block_real_robot_two_cams
3
+ library_name: lerobot
4
+ license: apache-2.0
5
+ model_name: diffusion
6
+ pipeline_tag: robotics
7
+ tags:
8
+ - lerobot
9
+ - robotics
10
+ - diffusion
11
+ ---
12
+
13
+ # Model Card for diffusion
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+ [Diffusion Policy](https://huggingface.co/papers/2303.04137) treats visuomotor control as a generative diffusion process, producing smooth, multi-step action trajectories that excel at contact-rich manipulation.
19
+
20
+
21
+ This policy has been trained and pushed to the Hub using [LeRobot](https://github.com/huggingface/lerobot).
22
+ See the full documentation at [LeRobot Docs](https://huggingface.co/docs/lerobot/index).
23
+
24
+ ---
25
+
26
+ ## How to Get Started with the Model
27
+
28
+ For a complete walkthrough, see the [training guide](https://huggingface.co/docs/lerobot/il_robots#train-a-policy).
29
+ Below is the short version on how to train and run inference/eval:
30
+
31
+ ### Train from scratch
32
+
33
+ ```bash
34
+ lerobot-train \
35
+ --dataset.repo_id=${HF_USER}/<dataset> \
36
+ --policy.type=act \
37
+ --output_dir=outputs/train/<desired_policy_repo_id> \
38
+ --job_name=lerobot_training \
39
+ --policy.device=cuda \
40
+ --policy.repo_id=${HF_USER}/<desired_policy_repo_id>
41
+ --wandb.enable=true
42
+ ```
43
+
44
+ _Writes checkpoints to `outputs/train/<desired_policy_repo_id>/checkpoints/`._
45
+
46
+ ### Evaluate the policy/run inference
47
+
48
+ ```bash
49
+ lerobot-record \
50
+ --robot.type=so100_follower \
51
+ --dataset.repo_id=<hf_user>/eval_<dataset> \
52
+ --policy.path=<hf_user>/<desired_policy_repo_id> \
53
+ --episodes=10
54
+ ```
55
+
56
+ Prefix the dataset repo with **eval\_** and supply `--policy.path` pointing to a local or hub checkpoint.
57
+
58
+ ---
59
+
60
+ ## Model Details
61
+
62
+ - **License:** apache-2.0
config.json ADDED
@@ -0,0 +1,95 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "diffusion",
3
+ "n_obs_steps": 2,
4
+ "input_features": {
5
+ "observation.image.cam0": {
6
+ "type": "VISUAL",
7
+ "shape": [
8
+ 3,
9
+ 180,
10
+ 320
11
+ ]
12
+ },
13
+ "observation.image.cam1": {
14
+ "type": "VISUAL",
15
+ "shape": [
16
+ 3,
17
+ 180,
18
+ 320
19
+ ]
20
+ },
21
+ "observation.state": {
22
+ "type": "STATE",
23
+ "shape": [
24
+ 3
25
+ ]
26
+ }
27
+ },
28
+ "output_features": {
29
+ "action": {
30
+ "type": "ACTION",
31
+ "shape": [
32
+ 3
33
+ ]
34
+ }
35
+ },
36
+ "device": "cuda",
37
+ "use_amp": false,
38
+ "push_to_hub": true,
39
+ "repo_id": "inaas/push_block_real_robot_two_cams_fork",
40
+ "private": null,
41
+ "tags": null,
42
+ "license": null,
43
+ "pretrained_path": null,
44
+ "horizon": 16,
45
+ "n_action_steps": 8,
46
+ "normalization_mapping": {
47
+ "VISUAL": "MEAN_STD",
48
+ "STATE": "MIN_MAX",
49
+ "ACTION": "MIN_MAX"
50
+ },
51
+ "drop_n_last_frames": 7,
52
+ "vision_backbone": "resnet18",
53
+ "crop_shape": [
54
+ 162,
55
+ 288
56
+ ],
57
+ "crop_is_random": true,
58
+ "pretrained_backbone_weights": null,
59
+ "use_group_norm": true,
60
+ "spatial_softmax_num_keypoints": 32,
61
+ "use_separate_rgb_encoder_per_camera": true,
62
+ "use_vit_encoder": false,
63
+ "vit_embed_dim": 512,
64
+ "vit_depth": 6,
65
+ "vit_num_heads": 8,
66
+ "vit_dropout": 0.1,
67
+ "down_dims": [
68
+ 256,
69
+ 512,
70
+ 1024
71
+ ],
72
+ "kernel_size": 5,
73
+ "n_groups": 8,
74
+ "diffusion_step_embed_dim": 128,
75
+ "use_film_scale_modulation": true,
76
+ "noise_scheduler_type": "DDPM",
77
+ "num_train_timesteps": 100,
78
+ "beta_schedule": "squaredcos_cap_v2",
79
+ "beta_start": 0.0001,
80
+ "beta_end": 0.02,
81
+ "prediction_type": "epsilon",
82
+ "clip_sample": true,
83
+ "clip_sample_range": 1.0,
84
+ "num_inference_steps": 10,
85
+ "do_mask_loss_for_padding": false,
86
+ "optimizer_lr": 0.0001,
87
+ "optimizer_betas": [
88
+ 0.95,
89
+ 0.999
90
+ ],
91
+ "optimizer_eps": 1e-08,
92
+ "optimizer_weight_decay": 1e-06,
93
+ "scheduler_name": "cosine",
94
+ "scheduler_warmup_steps": 500
95
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d6ab51c322b07c50117fa4e210c8b7b3052a0e81247922158dca3907db570068
3
+ size 356565772
train_config.json ADDED
@@ -0,0 +1,229 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "inaas/push_block_real_robot_two_cams",
4
+ "root": null,
5
+ "episodes": null,
6
+ "image_transforms": {
7
+ "enable": true,
8
+ "max_num_transforms": 3,
9
+ "random_order": false,
10
+ "tfs": {
11
+ "brightness": {
12
+ "weight": 1.0,
13
+ "type": "ColorJitter",
14
+ "kwargs": {
15
+ "brightness": [
16
+ 0.8,
17
+ 1.2
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": 1.0,
43
+ "type": "ColorJitter",
44
+ "kwargs": {
45
+ "hue": [
46
+ -0.05,
47
+ 0.05
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": "diffusion",
85
+ "n_obs_steps": 2,
86
+ "input_features": {
87
+ "observation.image.cam0": {
88
+ "type": "VISUAL",
89
+ "shape": [
90
+ 3,
91
+ 180,
92
+ 320
93
+ ]
94
+ },
95
+ "observation.image.cam1": {
96
+ "type": "VISUAL",
97
+ "shape": [
98
+ 3,
99
+ 180,
100
+ 320
101
+ ]
102
+ },
103
+ "observation.state": {
104
+ "type": "STATE",
105
+ "shape": [
106
+ 3
107
+ ]
108
+ }
109
+ },
110
+ "output_features": {
111
+ "action": {
112
+ "type": "ACTION",
113
+ "shape": [
114
+ 3
115
+ ]
116
+ }
117
+ },
118
+ "device": "cuda",
119
+ "use_amp": false,
120
+ "push_to_hub": true,
121
+ "repo_id": "inaas/push_block_real_robot_two_cams_fork",
122
+ "private": null,
123
+ "tags": null,
124
+ "license": null,
125
+ "pretrained_path": null,
126
+ "horizon": 16,
127
+ "n_action_steps": 8,
128
+ "normalization_mapping": {
129
+ "VISUAL": "MEAN_STD",
130
+ "STATE": "MIN_MAX",
131
+ "ACTION": "MIN_MAX"
132
+ },
133
+ "drop_n_last_frames": 7,
134
+ "vision_backbone": "resnet18",
135
+ "crop_shape": [
136
+ 162,
137
+ 288
138
+ ],
139
+ "crop_is_random": true,
140
+ "pretrained_backbone_weights": null,
141
+ "use_group_norm": true,
142
+ "spatial_softmax_num_keypoints": 32,
143
+ "use_separate_rgb_encoder_per_camera": true,
144
+ "use_vit_encoder": false,
145
+ "vit_embed_dim": 512,
146
+ "vit_depth": 6,
147
+ "vit_num_heads": 8,
148
+ "vit_dropout": 0.1,
149
+ "down_dims": [
150
+ 256,
151
+ 512,
152
+ 1024
153
+ ],
154
+ "kernel_size": 5,
155
+ "n_groups": 8,
156
+ "diffusion_step_embed_dim": 128,
157
+ "use_film_scale_modulation": true,
158
+ "noise_scheduler_type": "DDPM",
159
+ "num_train_timesteps": 100,
160
+ "beta_schedule": "squaredcos_cap_v2",
161
+ "beta_start": 0.0001,
162
+ "beta_end": 0.02,
163
+ "prediction_type": "epsilon",
164
+ "clip_sample": true,
165
+ "clip_sample_range": 1.0,
166
+ "num_inference_steps": 10,
167
+ "do_mask_loss_for_padding": false,
168
+ "optimizer_lr": 0.0001,
169
+ "optimizer_betas": [
170
+ 0.95,
171
+ 0.999
172
+ ],
173
+ "optimizer_eps": 1e-08,
174
+ "optimizer_weight_decay": 1e-06,
175
+ "scheduler_name": "cosine",
176
+ "scheduler_warmup_steps": 500
177
+ },
178
+ "output_dir": "/scratch/inaas/inaas/runs/lerobot/8504723/checkpoints",
179
+ "job_name": "diffusion",
180
+ "resume": false,
181
+ "seed": 1000,
182
+ "num_workers": 4,
183
+ "batch_size": 8,
184
+ "steps": 100000,
185
+ "eval_freq": 20000,
186
+ "log_freq": 200,
187
+ "tolerance_s": 0.0001,
188
+ "save_checkpoint": true,
189
+ "save_freq": 20000,
190
+ "use_policy_training_preset": true,
191
+ "optimizer": {
192
+ "type": "adam",
193
+ "lr": 0.0001,
194
+ "weight_decay": 1e-06,
195
+ "grad_clip_norm": 10.0,
196
+ "betas": [
197
+ 0.95,
198
+ 0.999
199
+ ],
200
+ "eps": 1e-08
201
+ },
202
+ "scheduler": {
203
+ "type": "diffuser",
204
+ "num_warmup_steps": 500,
205
+ "name": "cosine"
206
+ },
207
+ "eval": {
208
+ "n_episodes": 50,
209
+ "batch_size": 50,
210
+ "use_async_envs": false
211
+ },
212
+ "wandb": {
213
+ "enable": true,
214
+ "disable_artifact": false,
215
+ "project": "lerobot",
216
+ "entity": null,
217
+ "notes": null,
218
+ "run_id": "mx19mfw0",
219
+ "mode": null
220
+ },
221
+ "use_rabc": false,
222
+ "rabc_progress_path": null,
223
+ "rabc_kappa": 0.01,
224
+ "rabc_epsilon": 1e-06,
225
+ "rabc_head_mode": "sparse",
226
+ "use_umi_relative_actions": false,
227
+ "rename_map": {},
228
+ "checkpoint_path": null
229
+ }