Upload 8 files
Browse files- pour/config.json +254 -0
- pour/pour.ckpt +3 -0
- scoop/config.json +264 -0
- scoop/scoop.ckpt +3 -0
- shoe_org/config.json +261 -0
- shoe_org/shoe_org.ckpt +3 -0
- toy_collect/config.json +261 -0
- toy_collect/toy_collect.ckpt +3 -0
pour/config.json
ADDED
|
@@ -0,0 +1,254 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"num_steps": 160000,
|
| 3 |
+
"eval_every": 100000,
|
| 4 |
+
"validate_every": 1000,
|
| 5 |
+
"save_every": 8000,
|
| 6 |
+
"resume_ckpt_path": null,
|
| 7 |
+
"lr": 0.0001,
|
| 8 |
+
"policy_class": "cross",
|
| 9 |
+
"onscreen_render": false,
|
| 10 |
+
"seed": 6,
|
| 11 |
+
"temporal_agg": false,
|
| 12 |
+
"real_robot": true,
|
| 13 |
+
"load_pretrain": false,
|
| 14 |
+
"height": 480,
|
| 15 |
+
"width": 1280,
|
| 16 |
+
"normalize_resnet": false,
|
| 17 |
+
"wandb": true,
|
| 18 |
+
"pretrained_path": null,
|
| 19 |
+
"randomize_data_degree": 3,
|
| 20 |
+
"randomize_data": false,
|
| 21 |
+
"config": null,
|
| 22 |
+
"eval": false,
|
| 23 |
+
"batch_size": 32,
|
| 24 |
+
"lr_tokenizer": 0.0001,
|
| 25 |
+
"lr_action_head": 0.0001,
|
| 26 |
+
"lr_trunk": 0.0001,
|
| 27 |
+
"skip_mirrored_data": false,
|
| 28 |
+
"actuator_network_dir": null,
|
| 29 |
+
"history_len": null,
|
| 30 |
+
"future_len": null,
|
| 31 |
+
"prediction_len": null,
|
| 32 |
+
"kl_weight": null,
|
| 33 |
+
"chunk_size": 180,
|
| 34 |
+
"hidden_dim": null,
|
| 35 |
+
"dim_feedforward": null,
|
| 36 |
+
"no_encoder": true,
|
| 37 |
+
"dec_layers": 3,
|
| 38 |
+
"nheads": 8,
|
| 39 |
+
"use_pos_embd_image": 1,
|
| 40 |
+
"use_pos_embd_action": 1,
|
| 41 |
+
"feature_loss_weight": 0.0,
|
| 42 |
+
"self_attention": 1,
|
| 43 |
+
"same_backbones": true,
|
| 44 |
+
"use_mask": false,
|
| 45 |
+
"use_wrist": true,
|
| 46 |
+
"data_aug": false,
|
| 47 |
+
"grayscale": false,
|
| 48 |
+
"randomize_color": false,
|
| 49 |
+
"pretrain_image_width": 1276,
|
| 50 |
+
"pretrain_image_height": 480,
|
| 51 |
+
"gpu_id": 0,
|
| 52 |
+
"train_ratio": 0.99,
|
| 53 |
+
"share_cross_attn": false,
|
| 54 |
+
"policy_config": {
|
| 55 |
+
"lr": 0.0001,
|
| 56 |
+
"lr_backbone": 1e-05,
|
| 57 |
+
"embodiment_args_dict": {
|
| 58 |
+
"normalize_state": true,
|
| 59 |
+
"locoman": {
|
| 60 |
+
"tokenizer": {
|
| 61 |
+
"hidden_dim": 128,
|
| 62 |
+
"nhead": 4,
|
| 63 |
+
"dim_head": 32,
|
| 64 |
+
"dropout": 0.1,
|
| 65 |
+
"main_image": {
|
| 66 |
+
"output_dim": 128,
|
| 67 |
+
"weights": "DEFAULT",
|
| 68 |
+
"resnet_model": "resnet18",
|
| 69 |
+
"num_of_copy": 1,
|
| 70 |
+
"token_num": 16
|
| 71 |
+
},
|
| 72 |
+
"body_pose_state": {
|
| 73 |
+
"input_dim": 6,
|
| 74 |
+
"output_dim": 128,
|
| 75 |
+
"widths": [
|
| 76 |
+
128,
|
| 77 |
+
128
|
| 78 |
+
],
|
| 79 |
+
"tanh_end": false,
|
| 80 |
+
"ln": true,
|
| 81 |
+
"token_num": 4
|
| 82 |
+
},
|
| 83 |
+
"eef_pose_state": {
|
| 84 |
+
"input_dim": 12,
|
| 85 |
+
"output_dim": 128,
|
| 86 |
+
"widths": [
|
| 87 |
+
128,
|
| 88 |
+
128
|
| 89 |
+
],
|
| 90 |
+
"tanh_end": false,
|
| 91 |
+
"ln": true,
|
| 92 |
+
"token_num": 4
|
| 93 |
+
},
|
| 94 |
+
"eef_to_body_pose_state": {
|
| 95 |
+
"input_dim": 12,
|
| 96 |
+
"output_dim": 128,
|
| 97 |
+
"widths": [
|
| 98 |
+
128,
|
| 99 |
+
128
|
| 100 |
+
],
|
| 101 |
+
"tanh_end": false,
|
| 102 |
+
"ln": true,
|
| 103 |
+
"token_num": 4
|
| 104 |
+
},
|
| 105 |
+
"gripper_state": {
|
| 106 |
+
"input_dim": 2,
|
| 107 |
+
"output_dim": 128,
|
| 108 |
+
"widths": [
|
| 109 |
+
128,
|
| 110 |
+
128
|
| 111 |
+
],
|
| 112 |
+
"tanh_end": false,
|
| 113 |
+
"ln": true,
|
| 114 |
+
"token_num": 4
|
| 115 |
+
}
|
| 116 |
+
},
|
| 117 |
+
"action_head": {
|
| 118 |
+
"eef_pose": {
|
| 119 |
+
"type": "transformer",
|
| 120 |
+
"input_dim": 128,
|
| 121 |
+
"output_dim": 12,
|
| 122 |
+
"crossattn_modality_dropout": 0.1,
|
| 123 |
+
"crossattn_heads": 4,
|
| 124 |
+
"crossattn_dim_head": 16,
|
| 125 |
+
"action_horizon": 180
|
| 126 |
+
},
|
| 127 |
+
"body_pose": {
|
| 128 |
+
"type": "transformer",
|
| 129 |
+
"input_dim": 128,
|
| 130 |
+
"output_dim": 6,
|
| 131 |
+
"crossattn_modality_dropout": 0.1,
|
| 132 |
+
"crossattn_heads": 4,
|
| 133 |
+
"crossattn_dim_head": 16,
|
| 134 |
+
"action_horizon": 180
|
| 135 |
+
},
|
| 136 |
+
"gripper": {
|
| 137 |
+
"type": "transformer",
|
| 138 |
+
"input_dim": 128,
|
| 139 |
+
"output_dim": 2,
|
| 140 |
+
"crossattn_modality_dropout": 0.1,
|
| 141 |
+
"crossattn_heads": 4,
|
| 142 |
+
"crossattn_dim_head": 16,
|
| 143 |
+
"action_horizon": 180
|
| 144 |
+
}
|
| 145 |
+
}
|
| 146 |
+
},
|
| 147 |
+
"human": {
|
| 148 |
+
"tokenizer": {
|
| 149 |
+
"hidden_dim": 128,
|
| 150 |
+
"nhead": 4,
|
| 151 |
+
"dim_head": 32,
|
| 152 |
+
"dropout": 0.1,
|
| 153 |
+
"main_image": {
|
| 154 |
+
"output_dim": 128,
|
| 155 |
+
"weights": "DEFAULT",
|
| 156 |
+
"resnet_model": "resnet18",
|
| 157 |
+
"num_of_copy": 1,
|
| 158 |
+
"token_num": 16
|
| 159 |
+
},
|
| 160 |
+
"body_pose_state": {
|
| 161 |
+
"input_dim": 6,
|
| 162 |
+
"output_dim": 128,
|
| 163 |
+
"widths": [
|
| 164 |
+
128,
|
| 165 |
+
128
|
| 166 |
+
],
|
| 167 |
+
"tanh_end": false,
|
| 168 |
+
"ln": true,
|
| 169 |
+
"token_num": 4
|
| 170 |
+
},
|
| 171 |
+
"eef_pose_state": {
|
| 172 |
+
"input_dim": 12,
|
| 173 |
+
"output_dim": 128,
|
| 174 |
+
"widths": [
|
| 175 |
+
128,
|
| 176 |
+
128
|
| 177 |
+
],
|
| 178 |
+
"tanh_end": false,
|
| 179 |
+
"ln": true,
|
| 180 |
+
"token_num": 4
|
| 181 |
+
},
|
| 182 |
+
"eef_to_body_pose_state": {
|
| 183 |
+
"input_dim": 12,
|
| 184 |
+
"output_dim": 128,
|
| 185 |
+
"widths": [
|
| 186 |
+
128,
|
| 187 |
+
128
|
| 188 |
+
],
|
| 189 |
+
"tanh_end": false,
|
| 190 |
+
"ln": true,
|
| 191 |
+
"token_num": 4
|
| 192 |
+
},
|
| 193 |
+
"gripper_state": {
|
| 194 |
+
"input_dim": 2,
|
| 195 |
+
"output_dim": 128,
|
| 196 |
+
"widths": [
|
| 197 |
+
128,
|
| 198 |
+
128
|
| 199 |
+
],
|
| 200 |
+
"tanh_end": false,
|
| 201 |
+
"ln": true,
|
| 202 |
+
"token_num": 4
|
| 203 |
+
}
|
| 204 |
+
},
|
| 205 |
+
"action_head": {
|
| 206 |
+
"eef_pose": {
|
| 207 |
+
"type": "transformer",
|
| 208 |
+
"input_dim": 128,
|
| 209 |
+
"output_dim": 12,
|
| 210 |
+
"crossattn_modality_dropout": 0.1,
|
| 211 |
+
"crossattn_heads": 4,
|
| 212 |
+
"crossattn_dim_head": 16,
|
| 213 |
+
"action_horizon": 180
|
| 214 |
+
},
|
| 215 |
+
"body_pose": {
|
| 216 |
+
"type": "transformer",
|
| 217 |
+
"input_dim": 128,
|
| 218 |
+
"output_dim": 6,
|
| 219 |
+
"crossattn_modality_dropout": 0.1,
|
| 220 |
+
"crossattn_heads": 4,
|
| 221 |
+
"crossattn_dim_head": 16,
|
| 222 |
+
"action_horizon": 180
|
| 223 |
+
},
|
| 224 |
+
"gripper": {
|
| 225 |
+
"type": "transformer",
|
| 226 |
+
"input_dim": 128,
|
| 227 |
+
"output_dim": 2,
|
| 228 |
+
"crossattn_modality_dropout": 0.1,
|
| 229 |
+
"crossattn_heads": 4,
|
| 230 |
+
"crossattn_dim_head": 16,
|
| 231 |
+
"action_horizon": 180
|
| 232 |
+
}
|
| 233 |
+
}
|
| 234 |
+
}
|
| 235 |
+
},
|
| 236 |
+
"transformer_args": {
|
| 237 |
+
"output_len": {
|
| 238 |
+
"eef_pose": 6,
|
| 239 |
+
"body_pose": 6,
|
| 240 |
+
"gripper": 6
|
| 241 |
+
},
|
| 242 |
+
"trunk_mode": "encoder_decoder",
|
| 243 |
+
"hidden_dim": 128,
|
| 244 |
+
"dropout": 0.4,
|
| 245 |
+
"nheads": 16,
|
| 246 |
+
"dim_feedforward": 256,
|
| 247 |
+
"enc_layers": 4,
|
| 248 |
+
"dec_layers": 4,
|
| 249 |
+
"pre_norm": false,
|
| 250 |
+
"token_postprocessing": "none"
|
| 251 |
+
},
|
| 252 |
+
"share_cross_attn": false
|
| 253 |
+
}
|
| 254 |
+
}
|
pour/pour.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:aaa15e5ffcc46d91f53df96d4445ee3b72dda6ab2a459d36257091d26f14434f
|
| 3 |
+
size 52447058
|
scoop/config.json
ADDED
|
@@ -0,0 +1,264 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"num_steps": 200000,
|
| 3 |
+
"eval_every": 100000,
|
| 4 |
+
"validate_every": 1000,
|
| 5 |
+
"save_every": 5000,
|
| 6 |
+
"resume_ckpt_path": null,
|
| 7 |
+
"lr": 0.0001,
|
| 8 |
+
"policy_class": "cross",
|
| 9 |
+
"onscreen_render": false,
|
| 10 |
+
"seed": 6,
|
| 11 |
+
"temporal_agg": false,
|
| 12 |
+
"real_robot": true,
|
| 13 |
+
"load_pretrain": false,
|
| 14 |
+
"height": 480,
|
| 15 |
+
"width": 1280,
|
| 16 |
+
"normalize_resnet": false,
|
| 17 |
+
"wandb": true,
|
| 18 |
+
"pretrained_path": null,
|
| 19 |
+
"randomize_data_degree": 3,
|
| 20 |
+
"randomize_data": false,
|
| 21 |
+
"config": null,
|
| 22 |
+
"eval": false,
|
| 23 |
+
"batch_size": 32,
|
| 24 |
+
"lr_tokenizer": 0.0001,
|
| 25 |
+
"lr_action_head": 0.0001,
|
| 26 |
+
"lr_trunk": 0.0001,
|
| 27 |
+
"skip_mirrored_data": false,
|
| 28 |
+
"actuator_network_dir": null,
|
| 29 |
+
"history_len": null,
|
| 30 |
+
"future_len": null,
|
| 31 |
+
"prediction_len": null,
|
| 32 |
+
"kl_weight": null,
|
| 33 |
+
"chunk_size": 120,
|
| 34 |
+
"hidden_dim": null,
|
| 35 |
+
"dim_feedforward": null,
|
| 36 |
+
"no_encoder": true,
|
| 37 |
+
"dec_layers": 3,
|
| 38 |
+
"nheads": 8,
|
| 39 |
+
"use_pos_embd_image": 1,
|
| 40 |
+
"use_pos_embd_action": 1,
|
| 41 |
+
"feature_loss_weight": 0.0,
|
| 42 |
+
"self_attention": 1,
|
| 43 |
+
"same_backbones": true,
|
| 44 |
+
"use_mask": false,
|
| 45 |
+
"use_wrist": true,
|
| 46 |
+
"data_aug": false,
|
| 47 |
+
"grayscale": false,
|
| 48 |
+
"randomize_color": false,
|
| 49 |
+
"pretrain_image_width": 1276,
|
| 50 |
+
"pretrain_image_height": 480,
|
| 51 |
+
"gpu_id": 0,
|
| 52 |
+
"train_ratio": 0.99,
|
| 53 |
+
"min_val_num": 20,
|
| 54 |
+
"share_cross_attn": false,
|
| 55 |
+
"agg_modalities": false,
|
| 56 |
+
"policy_config": {
|
| 57 |
+
"lr": 0.0001,
|
| 58 |
+
"lr_backbone": 1e-05,
|
| 59 |
+
"embodiment_args_dict": {
|
| 60 |
+
"normalize_state": true,
|
| 61 |
+
"locoman": {
|
| 62 |
+
"tokenizer": {
|
| 63 |
+
"hidden_dim": 128,
|
| 64 |
+
"nhead": 4,
|
| 65 |
+
"dim_head": 32,
|
| 66 |
+
"dropout": 0.1,
|
| 67 |
+
"main_image": {
|
| 68 |
+
"output_dim": 128,
|
| 69 |
+
"weights": "DEFAULT",
|
| 70 |
+
"resnet_model": "resnet18",
|
| 71 |
+
"num_of_copy": 1,
|
| 72 |
+
"token_num": 16
|
| 73 |
+
},
|
| 74 |
+
"wrist_image": {
|
| 75 |
+
"output_dim": 128,
|
| 76 |
+
"weights": "DEFAULT",
|
| 77 |
+
"resnet_model": "resnet18",
|
| 78 |
+
"num_of_copy": 1,
|
| 79 |
+
"token_num": 8
|
| 80 |
+
},
|
| 81 |
+
"body_pose_state": {
|
| 82 |
+
"input_dim": 6,
|
| 83 |
+
"output_dim": 128,
|
| 84 |
+
"widths": [
|
| 85 |
+
128,
|
| 86 |
+
128
|
| 87 |
+
],
|
| 88 |
+
"tanh_end": false,
|
| 89 |
+
"ln": true,
|
| 90 |
+
"token_num": 4
|
| 91 |
+
},
|
| 92 |
+
"eef_pose_state": {
|
| 93 |
+
"input_dim": 12,
|
| 94 |
+
"output_dim": 128,
|
| 95 |
+
"widths": [
|
| 96 |
+
128,
|
| 97 |
+
128
|
| 98 |
+
],
|
| 99 |
+
"tanh_end": false,
|
| 100 |
+
"ln": true,
|
| 101 |
+
"token_num": 4
|
| 102 |
+
},
|
| 103 |
+
"eef_to_body_pose_state": {
|
| 104 |
+
"input_dim": 12,
|
| 105 |
+
"output_dim": 128,
|
| 106 |
+
"widths": [
|
| 107 |
+
128,
|
| 108 |
+
128
|
| 109 |
+
],
|
| 110 |
+
"tanh_end": false,
|
| 111 |
+
"ln": true,
|
| 112 |
+
"token_num": 4
|
| 113 |
+
},
|
| 114 |
+
"gripper_state": {
|
| 115 |
+
"input_dim": 2,
|
| 116 |
+
"output_dim": 128,
|
| 117 |
+
"widths": [
|
| 118 |
+
128,
|
| 119 |
+
128
|
| 120 |
+
],
|
| 121 |
+
"tanh_end": false,
|
| 122 |
+
"ln": true,
|
| 123 |
+
"token_num": 4
|
| 124 |
+
}
|
| 125 |
+
},
|
| 126 |
+
"action_head": {
|
| 127 |
+
"eef_pose": {
|
| 128 |
+
"type": "transformer",
|
| 129 |
+
"input_dim": 128,
|
| 130 |
+
"output_dim": 12,
|
| 131 |
+
"crossattn_modality_dropout": 0.1,
|
| 132 |
+
"crossattn_heads": 4,
|
| 133 |
+
"crossattn_dim_head": 16,
|
| 134 |
+
"action_horizon": 120
|
| 135 |
+
},
|
| 136 |
+
"body_pose": {
|
| 137 |
+
"type": "transformer",
|
| 138 |
+
"input_dim": 128,
|
| 139 |
+
"output_dim": 6,
|
| 140 |
+
"crossattn_modality_dropout": 0.1,
|
| 141 |
+
"crossattn_heads": 4,
|
| 142 |
+
"crossattn_dim_head": 16,
|
| 143 |
+
"action_horizon": 120
|
| 144 |
+
},
|
| 145 |
+
"gripper": {
|
| 146 |
+
"type": "transformer",
|
| 147 |
+
"input_dim": 128,
|
| 148 |
+
"output_dim": 2,
|
| 149 |
+
"crossattn_modality_dropout": 0.1,
|
| 150 |
+
"crossattn_heads": 4,
|
| 151 |
+
"crossattn_dim_head": 16,
|
| 152 |
+
"action_horizon": 120
|
| 153 |
+
}
|
| 154 |
+
}
|
| 155 |
+
},
|
| 156 |
+
"human": {
|
| 157 |
+
"tokenizer": {
|
| 158 |
+
"hidden_dim": 128,
|
| 159 |
+
"nhead": 4,
|
| 160 |
+
"dim_head": 32,
|
| 161 |
+
"dropout": 0.1,
|
| 162 |
+
"main_image": {
|
| 163 |
+
"output_dim": 128,
|
| 164 |
+
"weights": "DEFAULT",
|
| 165 |
+
"resnet_model": "resnet18",
|
| 166 |
+
"num_of_copy": 1,
|
| 167 |
+
"token_num": 16
|
| 168 |
+
},
|
| 169 |
+
"body_pose_state": {
|
| 170 |
+
"input_dim": 6,
|
| 171 |
+
"output_dim": 128,
|
| 172 |
+
"widths": [
|
| 173 |
+
128,
|
| 174 |
+
128
|
| 175 |
+
],
|
| 176 |
+
"tanh_end": false,
|
| 177 |
+
"ln": true,
|
| 178 |
+
"token_num": 4
|
| 179 |
+
},
|
| 180 |
+
"eef_pose_state": {
|
| 181 |
+
"input_dim": 12,
|
| 182 |
+
"output_dim": 128,
|
| 183 |
+
"widths": [
|
| 184 |
+
128,
|
| 185 |
+
128
|
| 186 |
+
],
|
| 187 |
+
"tanh_end": false,
|
| 188 |
+
"ln": true,
|
| 189 |
+
"token_num": 4
|
| 190 |
+
},
|
| 191 |
+
"eef_to_body_pose_state": {
|
| 192 |
+
"input_dim": 12,
|
| 193 |
+
"output_dim": 128,
|
| 194 |
+
"widths": [
|
| 195 |
+
128,
|
| 196 |
+
128
|
| 197 |
+
],
|
| 198 |
+
"tanh_end": false,
|
| 199 |
+
"ln": true,
|
| 200 |
+
"token_num": 4
|
| 201 |
+
},
|
| 202 |
+
"gripper_state": {
|
| 203 |
+
"input_dim": 2,
|
| 204 |
+
"output_dim": 128,
|
| 205 |
+
"widths": [
|
| 206 |
+
128,
|
| 207 |
+
128
|
| 208 |
+
],
|
| 209 |
+
"tanh_end": false,
|
| 210 |
+
"ln": true,
|
| 211 |
+
"token_num": 4
|
| 212 |
+
}
|
| 213 |
+
},
|
| 214 |
+
"action_head": {
|
| 215 |
+
"eef_pose": {
|
| 216 |
+
"type": "transformer",
|
| 217 |
+
"input_dim": 128,
|
| 218 |
+
"output_dim": 12,
|
| 219 |
+
"crossattn_modality_dropout": 0.1,
|
| 220 |
+
"crossattn_heads": 4,
|
| 221 |
+
"crossattn_dim_head": 16,
|
| 222 |
+
"action_horizon": 120
|
| 223 |
+
},
|
| 224 |
+
"body_pose": {
|
| 225 |
+
"type": "transformer",
|
| 226 |
+
"input_dim": 128,
|
| 227 |
+
"output_dim": 6,
|
| 228 |
+
"crossattn_modality_dropout": 0.1,
|
| 229 |
+
"crossattn_heads": 4,
|
| 230 |
+
"crossattn_dim_head": 16,
|
| 231 |
+
"action_horizon": 120
|
| 232 |
+
},
|
| 233 |
+
"gripper": {
|
| 234 |
+
"type": "transformer",
|
| 235 |
+
"input_dim": 128,
|
| 236 |
+
"output_dim": 2,
|
| 237 |
+
"crossattn_modality_dropout": 0.1,
|
| 238 |
+
"crossattn_heads": 4,
|
| 239 |
+
"crossattn_dim_head": 16,
|
| 240 |
+
"action_horizon": 120
|
| 241 |
+
}
|
| 242 |
+
}
|
| 243 |
+
}
|
| 244 |
+
},
|
| 245 |
+
"transformer_args": {
|
| 246 |
+
"output_len": {
|
| 247 |
+
"eef_pose": 6,
|
| 248 |
+
"body_pose": 6,
|
| 249 |
+
"gripper": 6
|
| 250 |
+
},
|
| 251 |
+
"trunk_mode": "encoder_decoder",
|
| 252 |
+
"hidden_dim": 128,
|
| 253 |
+
"dropout": 0.5,
|
| 254 |
+
"nheads": 16,
|
| 255 |
+
"dim_feedforward": 256,
|
| 256 |
+
"enc_layers": 4,
|
| 257 |
+
"dec_layers": 4,
|
| 258 |
+
"pre_norm": false,
|
| 259 |
+
"token_postprocessing": "none"
|
| 260 |
+
},
|
| 261 |
+
"share_cross_attn": false,
|
| 262 |
+
"agg_modalities": false
|
| 263 |
+
}
|
| 264 |
+
}
|
scoop/scoop.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:494865bb48acddb4542a4e0a91a5f830a7e9064aa6b4e1fe81c6a9dbff2abe39
|
| 3 |
+
size 52442322
|
shoe_org/config.json
ADDED
|
@@ -0,0 +1,261 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"num_steps": 160000,
|
| 3 |
+
"eval_every": 100000,
|
| 4 |
+
"validate_every": 1000,
|
| 5 |
+
"save_every": 8000,
|
| 6 |
+
"resume_ckpt_path": null,
|
| 7 |
+
"lr": 0.0001,
|
| 8 |
+
"policy_class": "cross",
|
| 9 |
+
"onscreen_render": false,
|
| 10 |
+
"seed": 6,
|
| 11 |
+
"temporal_agg": false,
|
| 12 |
+
"real_robot": true,
|
| 13 |
+
"load_pretrain": false,
|
| 14 |
+
"height": 480,
|
| 15 |
+
"width": 1280,
|
| 16 |
+
"normalize_resnet": false,
|
| 17 |
+
"wandb": true,
|
| 18 |
+
"pretrained_path": null,
|
| 19 |
+
"randomize_data_degree": 3,
|
| 20 |
+
"randomize_data": false,
|
| 21 |
+
"config": null,
|
| 22 |
+
"eval": false,
|
| 23 |
+
"batch_size": 32,
|
| 24 |
+
"lr_tokenizer": 0.0001,
|
| 25 |
+
"lr_action_head": 0.0001,
|
| 26 |
+
"lr_trunk": 0.0001,
|
| 27 |
+
"skip_mirrored_data": false,
|
| 28 |
+
"actuator_network_dir": null,
|
| 29 |
+
"history_len": null,
|
| 30 |
+
"future_len": null,
|
| 31 |
+
"prediction_len": null,
|
| 32 |
+
"kl_weight": null,
|
| 33 |
+
"chunk_size": 180,
|
| 34 |
+
"hidden_dim": null,
|
| 35 |
+
"dim_feedforward": null,
|
| 36 |
+
"no_encoder": true,
|
| 37 |
+
"dec_layers": 3,
|
| 38 |
+
"nheads": 8,
|
| 39 |
+
"use_pos_embd_image": 1,
|
| 40 |
+
"use_pos_embd_action": 1,
|
| 41 |
+
"feature_loss_weight": 0.0,
|
| 42 |
+
"self_attention": 1,
|
| 43 |
+
"same_backbones": true,
|
| 44 |
+
"use_mask": false,
|
| 45 |
+
"use_wrist": true,
|
| 46 |
+
"data_aug": false,
|
| 47 |
+
"grayscale": false,
|
| 48 |
+
"randomize_color": false,
|
| 49 |
+
"pretrain_image_width": 1276,
|
| 50 |
+
"pretrain_image_height": 480,
|
| 51 |
+
"gpu_id": 0,
|
| 52 |
+
"train_ratio": 0.99,
|
| 53 |
+
"share_cross_attn": false,
|
| 54 |
+
"policy_config": {
|
| 55 |
+
"lr": 0.0001,
|
| 56 |
+
"lr_backbone": 1e-05,
|
| 57 |
+
"embodiment_args_dict": {
|
| 58 |
+
"normalize_state": true,
|
| 59 |
+
"locoman": {
|
| 60 |
+
"tokenizer": {
|
| 61 |
+
"hidden_dim": 128,
|
| 62 |
+
"nhead": 4,
|
| 63 |
+
"dim_head": 32,
|
| 64 |
+
"dropout": 0.1,
|
| 65 |
+
"main_image": {
|
| 66 |
+
"output_dim": 128,
|
| 67 |
+
"weights": "DEFAULT",
|
| 68 |
+
"resnet_model": "resnet18",
|
| 69 |
+
"num_of_copy": 1,
|
| 70 |
+
"token_num": 16
|
| 71 |
+
},
|
| 72 |
+
"wrist_image": {
|
| 73 |
+
"output_dim": 128,
|
| 74 |
+
"weights": "DEFAULT",
|
| 75 |
+
"resnet_model": "resnet18",
|
| 76 |
+
"num_of_copy": 1,
|
| 77 |
+
"token_num": 8
|
| 78 |
+
},
|
| 79 |
+
"body_pose_state": {
|
| 80 |
+
"input_dim": 6,
|
| 81 |
+
"output_dim": 128,
|
| 82 |
+
"widths": [
|
| 83 |
+
128,
|
| 84 |
+
128
|
| 85 |
+
],
|
| 86 |
+
"tanh_end": false,
|
| 87 |
+
"ln": true,
|
| 88 |
+
"token_num": 4
|
| 89 |
+
},
|
| 90 |
+
"eef_pose_state": {
|
| 91 |
+
"input_dim": 12,
|
| 92 |
+
"output_dim": 128,
|
| 93 |
+
"widths": [
|
| 94 |
+
128,
|
| 95 |
+
128
|
| 96 |
+
],
|
| 97 |
+
"tanh_end": false,
|
| 98 |
+
"ln": true,
|
| 99 |
+
"token_num": 4
|
| 100 |
+
},
|
| 101 |
+
"eef_to_body_pose_state": {
|
| 102 |
+
"input_dim": 12,
|
| 103 |
+
"output_dim": 128,
|
| 104 |
+
"widths": [
|
| 105 |
+
128,
|
| 106 |
+
128
|
| 107 |
+
],
|
| 108 |
+
"tanh_end": false,
|
| 109 |
+
"ln": true,
|
| 110 |
+
"token_num": 4
|
| 111 |
+
},
|
| 112 |
+
"gripper_state": {
|
| 113 |
+
"input_dim": 2,
|
| 114 |
+
"output_dim": 128,
|
| 115 |
+
"widths": [
|
| 116 |
+
128,
|
| 117 |
+
128
|
| 118 |
+
],
|
| 119 |
+
"tanh_end": false,
|
| 120 |
+
"ln": true,
|
| 121 |
+
"token_num": 4
|
| 122 |
+
}
|
| 123 |
+
},
|
| 124 |
+
"action_head": {
|
| 125 |
+
"eef_pose": {
|
| 126 |
+
"type": "transformer",
|
| 127 |
+
"input_dim": 128,
|
| 128 |
+
"output_dim": 12,
|
| 129 |
+
"crossattn_modality_dropout": 0.1,
|
| 130 |
+
"crossattn_heads": 4,
|
| 131 |
+
"crossattn_dim_head": 16,
|
| 132 |
+
"action_horizon": 180
|
| 133 |
+
},
|
| 134 |
+
"body_pose": {
|
| 135 |
+
"type": "transformer",
|
| 136 |
+
"input_dim": 128,
|
| 137 |
+
"output_dim": 6,
|
| 138 |
+
"crossattn_modality_dropout": 0.1,
|
| 139 |
+
"crossattn_heads": 4,
|
| 140 |
+
"crossattn_dim_head": 16,
|
| 141 |
+
"action_horizon": 180
|
| 142 |
+
},
|
| 143 |
+
"gripper": {
|
| 144 |
+
"type": "transformer",
|
| 145 |
+
"input_dim": 128,
|
| 146 |
+
"output_dim": 2,
|
| 147 |
+
"crossattn_modality_dropout": 0.1,
|
| 148 |
+
"crossattn_heads": 4,
|
| 149 |
+
"crossattn_dim_head": 16,
|
| 150 |
+
"action_horizon": 180
|
| 151 |
+
}
|
| 152 |
+
}
|
| 153 |
+
},
|
| 154 |
+
"human": {
|
| 155 |
+
"tokenizer": {
|
| 156 |
+
"hidden_dim": 128,
|
| 157 |
+
"nhead": 4,
|
| 158 |
+
"dim_head": 32,
|
| 159 |
+
"dropout": 0.1,
|
| 160 |
+
"main_image": {
|
| 161 |
+
"output_dim": 128,
|
| 162 |
+
"weights": "DEFAULT",
|
| 163 |
+
"resnet_model": "resnet18",
|
| 164 |
+
"num_of_copy": 1,
|
| 165 |
+
"token_num": 16
|
| 166 |
+
},
|
| 167 |
+
"body_pose_state": {
|
| 168 |
+
"input_dim": 6,
|
| 169 |
+
"output_dim": 128,
|
| 170 |
+
"widths": [
|
| 171 |
+
128,
|
| 172 |
+
128
|
| 173 |
+
],
|
| 174 |
+
"tanh_end": false,
|
| 175 |
+
"ln": true,
|
| 176 |
+
"token_num": 4
|
| 177 |
+
},
|
| 178 |
+
"eef_pose_state": {
|
| 179 |
+
"input_dim": 12,
|
| 180 |
+
"output_dim": 128,
|
| 181 |
+
"widths": [
|
| 182 |
+
128,
|
| 183 |
+
128
|
| 184 |
+
],
|
| 185 |
+
"tanh_end": false,
|
| 186 |
+
"ln": true,
|
| 187 |
+
"token_num": 4
|
| 188 |
+
},
|
| 189 |
+
"eef_to_body_pose_state": {
|
| 190 |
+
"input_dim": 12,
|
| 191 |
+
"output_dim": 128,
|
| 192 |
+
"widths": [
|
| 193 |
+
128,
|
| 194 |
+
128
|
| 195 |
+
],
|
| 196 |
+
"tanh_end": false,
|
| 197 |
+
"ln": true,
|
| 198 |
+
"token_num": 4
|
| 199 |
+
},
|
| 200 |
+
"gripper_state": {
|
| 201 |
+
"input_dim": 2,
|
| 202 |
+
"output_dim": 128,
|
| 203 |
+
"widths": [
|
| 204 |
+
128,
|
| 205 |
+
128
|
| 206 |
+
],
|
| 207 |
+
"tanh_end": false,
|
| 208 |
+
"ln": true,
|
| 209 |
+
"token_num": 4
|
| 210 |
+
}
|
| 211 |
+
},
|
| 212 |
+
"action_head": {
|
| 213 |
+
"eef_pose": {
|
| 214 |
+
"type": "transformer",
|
| 215 |
+
"input_dim": 128,
|
| 216 |
+
"output_dim": 12,
|
| 217 |
+
"crossattn_modality_dropout": 0.1,
|
| 218 |
+
"crossattn_heads": 4,
|
| 219 |
+
"crossattn_dim_head": 16,
|
| 220 |
+
"action_horizon": 180
|
| 221 |
+
},
|
| 222 |
+
"body_pose": {
|
| 223 |
+
"type": "transformer",
|
| 224 |
+
"input_dim": 128,
|
| 225 |
+
"output_dim": 6,
|
| 226 |
+
"crossattn_modality_dropout": 0.1,
|
| 227 |
+
"crossattn_heads": 4,
|
| 228 |
+
"crossattn_dim_head": 16,
|
| 229 |
+
"action_horizon": 180
|
| 230 |
+
},
|
| 231 |
+
"gripper": {
|
| 232 |
+
"type": "transformer",
|
| 233 |
+
"input_dim": 128,
|
| 234 |
+
"output_dim": 2,
|
| 235 |
+
"crossattn_modality_dropout": 0.1,
|
| 236 |
+
"crossattn_heads": 4,
|
| 237 |
+
"crossattn_dim_head": 16,
|
| 238 |
+
"action_horizon": 180
|
| 239 |
+
}
|
| 240 |
+
}
|
| 241 |
+
}
|
| 242 |
+
},
|
| 243 |
+
"transformer_args": {
|
| 244 |
+
"output_len": {
|
| 245 |
+
"eef_pose": 6,
|
| 246 |
+
"body_pose": 6,
|
| 247 |
+
"gripper": 6
|
| 248 |
+
},
|
| 249 |
+
"trunk_mode": "encoder_decoder",
|
| 250 |
+
"hidden_dim": 128,
|
| 251 |
+
"dropout": 0.4,
|
| 252 |
+
"nheads": 16,
|
| 253 |
+
"dim_feedforward": 256,
|
| 254 |
+
"enc_layers": 4,
|
| 255 |
+
"dec_layers": 4,
|
| 256 |
+
"pre_norm": false,
|
| 257 |
+
"token_postprocessing": "none"
|
| 258 |
+
},
|
| 259 |
+
"share_cross_attn": false
|
| 260 |
+
}
|
| 261 |
+
}
|
shoe_org/shoe_org.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fd3d0787aaaeefded91d95f743abcf190abe22d485d39bda3e69c14c17259c9f
|
| 3 |
+
size 52451898
|
toy_collect/config.json
ADDED
|
@@ -0,0 +1,261 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"num_steps": 160000,
|
| 3 |
+
"eval_every": 100000,
|
| 4 |
+
"validate_every": 1000,
|
| 5 |
+
"save_every": 8000,
|
| 6 |
+
"resume_ckpt_path": null,
|
| 7 |
+
"lr": 0.0001,
|
| 8 |
+
"policy_class": "cross",
|
| 9 |
+
"onscreen_render": false,
|
| 10 |
+
"seed": 6,
|
| 11 |
+
"temporal_agg": false,
|
| 12 |
+
"real_robot": true,
|
| 13 |
+
"load_pretrain": false,
|
| 14 |
+
"height": 480,
|
| 15 |
+
"width": 1280,
|
| 16 |
+
"normalize_resnet": false,
|
| 17 |
+
"wandb": true,
|
| 18 |
+
"pretrained_path": null,
|
| 19 |
+
"randomize_data_degree": 3,
|
| 20 |
+
"randomize_data": false,
|
| 21 |
+
"config": null,
|
| 22 |
+
"eval": false,
|
| 23 |
+
"batch_size": 32,
|
| 24 |
+
"lr_tokenizer": 0.0001,
|
| 25 |
+
"lr_action_head": 0.0001,
|
| 26 |
+
"lr_trunk": 0.0001,
|
| 27 |
+
"skip_mirrored_data": false,
|
| 28 |
+
"actuator_network_dir": null,
|
| 29 |
+
"history_len": null,
|
| 30 |
+
"future_len": null,
|
| 31 |
+
"prediction_len": null,
|
| 32 |
+
"kl_weight": null,
|
| 33 |
+
"chunk_size": 60,
|
| 34 |
+
"hidden_dim": null,
|
| 35 |
+
"dim_feedforward": null,
|
| 36 |
+
"no_encoder": true,
|
| 37 |
+
"dec_layers": 3,
|
| 38 |
+
"nheads": 8,
|
| 39 |
+
"use_pos_embd_image": 1,
|
| 40 |
+
"use_pos_embd_action": 1,
|
| 41 |
+
"feature_loss_weight": 0.0,
|
| 42 |
+
"self_attention": 1,
|
| 43 |
+
"same_backbones": true,
|
| 44 |
+
"use_mask": false,
|
| 45 |
+
"use_wrist": true,
|
| 46 |
+
"data_aug": false,
|
| 47 |
+
"grayscale": false,
|
| 48 |
+
"randomize_color": false,
|
| 49 |
+
"pretrain_image_width": 1276,
|
| 50 |
+
"pretrain_image_height": 480,
|
| 51 |
+
"gpu_id": 0,
|
| 52 |
+
"train_ratio": 0.99,
|
| 53 |
+
"share_cross_attn": false,
|
| 54 |
+
"policy_config": {
|
| 55 |
+
"lr": 0.0001,
|
| 56 |
+
"lr_backbone": 1e-05,
|
| 57 |
+
"embodiment_args_dict": {
|
| 58 |
+
"normalize_state": true,
|
| 59 |
+
"locoman": {
|
| 60 |
+
"tokenizer": {
|
| 61 |
+
"hidden_dim": 128,
|
| 62 |
+
"nhead": 4,
|
| 63 |
+
"dim_head": 32,
|
| 64 |
+
"dropout": 0.1,
|
| 65 |
+
"main_image": {
|
| 66 |
+
"output_dim": 128,
|
| 67 |
+
"weights": "DEFAULT",
|
| 68 |
+
"resnet_model": "resnet18",
|
| 69 |
+
"num_of_copy": 1,
|
| 70 |
+
"token_num": 16
|
| 71 |
+
},
|
| 72 |
+
"wrist_image": {
|
| 73 |
+
"output_dim": 128,
|
| 74 |
+
"weights": "DEFAULT",
|
| 75 |
+
"resnet_model": "resnet18",
|
| 76 |
+
"num_of_copy": 1,
|
| 77 |
+
"token_num": 8
|
| 78 |
+
},
|
| 79 |
+
"body_pose_state": {
|
| 80 |
+
"input_dim": 6,
|
| 81 |
+
"output_dim": 128,
|
| 82 |
+
"widths": [
|
| 83 |
+
128,
|
| 84 |
+
128
|
| 85 |
+
],
|
| 86 |
+
"tanh_end": false,
|
| 87 |
+
"ln": true,
|
| 88 |
+
"token_num": 4
|
| 89 |
+
},
|
| 90 |
+
"eef_pose_state": {
|
| 91 |
+
"input_dim": 12,
|
| 92 |
+
"output_dim": 128,
|
| 93 |
+
"widths": [
|
| 94 |
+
128,
|
| 95 |
+
128
|
| 96 |
+
],
|
| 97 |
+
"tanh_end": false,
|
| 98 |
+
"ln": true,
|
| 99 |
+
"token_num": 4
|
| 100 |
+
},
|
| 101 |
+
"eef_to_body_pose_state": {
|
| 102 |
+
"input_dim": 12,
|
| 103 |
+
"output_dim": 128,
|
| 104 |
+
"widths": [
|
| 105 |
+
128,
|
| 106 |
+
128
|
| 107 |
+
],
|
| 108 |
+
"tanh_end": false,
|
| 109 |
+
"ln": true,
|
| 110 |
+
"token_num": 4
|
| 111 |
+
},
|
| 112 |
+
"gripper_state": {
|
| 113 |
+
"input_dim": 2,
|
| 114 |
+
"output_dim": 128,
|
| 115 |
+
"widths": [
|
| 116 |
+
128,
|
| 117 |
+
128
|
| 118 |
+
],
|
| 119 |
+
"tanh_end": false,
|
| 120 |
+
"ln": true,
|
| 121 |
+
"token_num": 4
|
| 122 |
+
}
|
| 123 |
+
},
|
| 124 |
+
"action_head": {
|
| 125 |
+
"eef_pose": {
|
| 126 |
+
"type": "transformer",
|
| 127 |
+
"input_dim": 128,
|
| 128 |
+
"output_dim": 12,
|
| 129 |
+
"crossattn_modality_dropout": 0.1,
|
| 130 |
+
"crossattn_heads": 4,
|
| 131 |
+
"crossattn_dim_head": 16,
|
| 132 |
+
"action_horizon": 60
|
| 133 |
+
},
|
| 134 |
+
"body_pose": {
|
| 135 |
+
"type": "transformer",
|
| 136 |
+
"input_dim": 128,
|
| 137 |
+
"output_dim": 6,
|
| 138 |
+
"crossattn_modality_dropout": 0.1,
|
| 139 |
+
"crossattn_heads": 4,
|
| 140 |
+
"crossattn_dim_head": 16,
|
| 141 |
+
"action_horizon": 60
|
| 142 |
+
},
|
| 143 |
+
"gripper": {
|
| 144 |
+
"type": "transformer",
|
| 145 |
+
"input_dim": 128,
|
| 146 |
+
"output_dim": 2,
|
| 147 |
+
"crossattn_modality_dropout": 0.1,
|
| 148 |
+
"crossattn_heads": 4,
|
| 149 |
+
"crossattn_dim_head": 16,
|
| 150 |
+
"action_horizon": 60
|
| 151 |
+
}
|
| 152 |
+
}
|
| 153 |
+
},
|
| 154 |
+
"human": {
|
| 155 |
+
"tokenizer": {
|
| 156 |
+
"hidden_dim": 128,
|
| 157 |
+
"nhead": 4,
|
| 158 |
+
"dim_head": 32,
|
| 159 |
+
"dropout": 0.1,
|
| 160 |
+
"main_image": {
|
| 161 |
+
"output_dim": 128,
|
| 162 |
+
"weights": "DEFAULT",
|
| 163 |
+
"resnet_model": "resnet18",
|
| 164 |
+
"num_of_copy": 1,
|
| 165 |
+
"token_num": 16
|
| 166 |
+
},
|
| 167 |
+
"body_pose_state": {
|
| 168 |
+
"input_dim": 6,
|
| 169 |
+
"output_dim": 128,
|
| 170 |
+
"widths": [
|
| 171 |
+
128,
|
| 172 |
+
128
|
| 173 |
+
],
|
| 174 |
+
"tanh_end": false,
|
| 175 |
+
"ln": true,
|
| 176 |
+
"token_num": 4
|
| 177 |
+
},
|
| 178 |
+
"eef_pose_state": {
|
| 179 |
+
"input_dim": 12,
|
| 180 |
+
"output_dim": 128,
|
| 181 |
+
"widths": [
|
| 182 |
+
128,
|
| 183 |
+
128
|
| 184 |
+
],
|
| 185 |
+
"tanh_end": false,
|
| 186 |
+
"ln": true,
|
| 187 |
+
"token_num": 4
|
| 188 |
+
},
|
| 189 |
+
"eef_to_body_pose_state": {
|
| 190 |
+
"input_dim": 12,
|
| 191 |
+
"output_dim": 128,
|
| 192 |
+
"widths": [
|
| 193 |
+
128,
|
| 194 |
+
128
|
| 195 |
+
],
|
| 196 |
+
"tanh_end": false,
|
| 197 |
+
"ln": true,
|
| 198 |
+
"token_num": 4
|
| 199 |
+
},
|
| 200 |
+
"gripper_state": {
|
| 201 |
+
"input_dim": 2,
|
| 202 |
+
"output_dim": 128,
|
| 203 |
+
"widths": [
|
| 204 |
+
128,
|
| 205 |
+
128
|
| 206 |
+
],
|
| 207 |
+
"tanh_end": false,
|
| 208 |
+
"ln": true,
|
| 209 |
+
"token_num": 4
|
| 210 |
+
}
|
| 211 |
+
},
|
| 212 |
+
"action_head": {
|
| 213 |
+
"eef_pose": {
|
| 214 |
+
"type": "transformer",
|
| 215 |
+
"input_dim": 128,
|
| 216 |
+
"output_dim": 12,
|
| 217 |
+
"crossattn_modality_dropout": 0.1,
|
| 218 |
+
"crossattn_heads": 4,
|
| 219 |
+
"crossattn_dim_head": 16,
|
| 220 |
+
"action_horizon": 60
|
| 221 |
+
},
|
| 222 |
+
"body_pose": {
|
| 223 |
+
"type": "transformer",
|
| 224 |
+
"input_dim": 128,
|
| 225 |
+
"output_dim": 6,
|
| 226 |
+
"crossattn_modality_dropout": 0.1,
|
| 227 |
+
"crossattn_heads": 4,
|
| 228 |
+
"crossattn_dim_head": 16,
|
| 229 |
+
"action_horizon": 60
|
| 230 |
+
},
|
| 231 |
+
"gripper": {
|
| 232 |
+
"type": "transformer",
|
| 233 |
+
"input_dim": 128,
|
| 234 |
+
"output_dim": 2,
|
| 235 |
+
"crossattn_modality_dropout": 0.1,
|
| 236 |
+
"crossattn_heads": 4,
|
| 237 |
+
"crossattn_dim_head": 16,
|
| 238 |
+
"action_horizon": 60
|
| 239 |
+
}
|
| 240 |
+
}
|
| 241 |
+
}
|
| 242 |
+
},
|
| 243 |
+
"transformer_args": {
|
| 244 |
+
"output_len": {
|
| 245 |
+
"eef_pose": 6,
|
| 246 |
+
"body_pose": 6,
|
| 247 |
+
"gripper": 6
|
| 248 |
+
},
|
| 249 |
+
"trunk_mode": "encoder_decoder",
|
| 250 |
+
"hidden_dim": 128,
|
| 251 |
+
"dropout": 0.4,
|
| 252 |
+
"nheads": 16,
|
| 253 |
+
"dim_feedforward": 256,
|
| 254 |
+
"enc_layers": 4,
|
| 255 |
+
"dec_layers": 4,
|
| 256 |
+
"pre_norm": false,
|
| 257 |
+
"token_postprocessing": "none"
|
| 258 |
+
},
|
| 259 |
+
"share_cross_attn": false
|
| 260 |
+
}
|
| 261 |
+
}
|
toy_collect/toy_collect.ckpt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0b183e4bfa0db8efeff596bd3f122ae8f45657aa756f67d7e83a5b928a14bcd6
|
| 3 |
+
size 52441829
|