Repository navigation
Expand file tree
/
Copy pathact.patch
More file actions
610 lines (550 loc) · 29.4 KB
/
Copy pathact.patch
File metadata and controls
610 lines (550 loc) · 29.4 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
diff --git a/assets/bimanual_viperx_transfer_cube_nuisance.xml b/assets/bimanual_viperx_transfer_cube_nuisance.xml
new file mode 100644
index 0000000..f4f0870
--- /dev/null
+++ b/assets/bimanual_viperx_transfer_cube_nuisance.xml
@@ -0,0 +1,53 @@
+<mujoco>
+ <!-- cfnbc: copy of bimanual_viperx_transfer_cube.xml + hidden nuisance geoms. Physics must match the original. -->
+ <include file="scene.xml"/>
+ <include file="vx300s_dependencies.xml"/>
+ <worldbody>
+ <include file="vx300s_left.xml" />
+ <include file="vx300s_right.xml" />
+
+ <body name="box" pos="0.2 0.5 0.05">
+ <joint name="red_box_joint" type="free" frictionloss="0.01" />
+ <inertial pos="0 0 0" mass="0.05" diaginertia="0.002 0.002 0.002" />
+ <geom condim="4" solimp="2 1 0.01" solref="0.01 1" friction="1 0.005 0.0001" pos="0 0 0" size="0.02 0.02 0.02" type="box" name="red_box" rgba="1 0 0 1" />
+ </body>
+
+ <!-- cfnbc nuisance geoms: visual only (contype/conaffinity 0, static bodies), hidden in geom group 4
+ until cfnbc.nuisance.apply() moves them to group 1 and places them relative to the cube. -->
+ <body name="nuis_support" pos="0 0.5 0">
+ <geom name="nuis_support" type="box" size="0.05 0.05 0.0005" pos="0 0 0.0006" rgba="0.45 0.45 0.45 1" group="4" contype="0" conaffinity="0" />
+ </body>
+ <body name="nuis_distractor_0" pos="0 0.5 0"><geom name="nuis_distractor_0" type="box" size="0.012 0.012 0.012" pos="0 0 0.012" rgba="0.1 0.6 1 1" group="4" contype="0" conaffinity="0" /></body>
+ <body name="nuis_distractor_1" pos="0 0.5 0"><geom name="nuis_distractor_1" type="box" size="0.012 0.012 0.012" pos="0 0 0.012" rgba="0.1 0.6 1 1" group="4" contype="0" conaffinity="0" /></body>
+ <body name="nuis_distractor_2" pos="0 0.5 0"><geom name="nuis_distractor_2" type="box" size="0.012 0.012 0.012" pos="0 0 0.012" rgba="0.1 0.6 1 1" group="4" contype="0" conaffinity="0" /></body>
+ <body name="nuis_distractor_3" pos="0 0.5 0"><geom name="nuis_distractor_3" type="box" size="0.012 0.012 0.012" pos="0 0 0.012" rgba="0.1 0.6 1 1" group="4" contype="0" conaffinity="0" /></body>
+
+ </worldbody>
+
+ <actuator>
+ <position ctrllimited="true" ctrlrange="-3.14158 3.14158" joint="vx300s_left/waist" kp="800" user="1" forcelimited="true" forcerange="-150 150"/>
+ <position ctrllimited="true" ctrlrange="-1.85005 1.25664" joint="vx300s_left/shoulder" kp="1600" user="1" forcelimited="true" forcerange="-300 300"/>
+ <position ctrllimited="true" ctrlrange="-1.76278 1.6057" joint="vx300s_left/elbow" kp="800" user="1" forcelimited="true" forcerange="-100 100"/>
+ <position ctrllimited="true" ctrlrange="-3.14158 3.14158" joint="vx300s_left/forearm_roll" kp="10" user="1" forcelimited="true" forcerange="-100 100"/>
+ <position ctrllimited="true" ctrlrange="-1.8675 2.23402" joint="vx300s_left/wrist_angle" kp="50" user="1"/>
+ <position ctrllimited="true" ctrlrange="-3.14158 3.14158" joint="vx300s_left/wrist_rotate" kp="20" user="1"/>
+ <position ctrllimited="true" ctrlrange="0.021 0.057" joint="vx300s_left/left_finger" kp="200" user="1"/>
+ <position ctrllimited="true" ctrlrange="-0.057 -0.021" joint="vx300s_left/right_finger" kp="200" user="1"/>
+
+ <position ctrllimited="true" ctrlrange="-3.14158 3.14158" joint="vx300s_right/waist" kp="800" user="1" forcelimited="true" forcerange="-150 150"/>
+ <position ctrllimited="true" ctrlrange="-1.85005 1.25664" joint="vx300s_right/shoulder" kp="1600" user="1" forcelimited="true" forcerange="-300 300"/>
+ <position ctrllimited="true" ctrlrange="-1.76278 1.6057" joint="vx300s_right/elbow" kp="800" user="1" forcelimited="true" forcerange="-100 100"/>
+ <position ctrllimited="true" ctrlrange="-3.14158 3.14158" joint="vx300s_right/forearm_roll" kp="10" user="1" forcelimited="true" forcerange="-100 100"/>
+ <position ctrllimited="true" ctrlrange="-1.8675 2.23402" joint="vx300s_right/wrist_angle" kp="50" user="1"/>
+ <position ctrllimited="true" ctrlrange="-3.14158 3.14158" joint="vx300s_right/wrist_rotate" kp="20" user="1"/>
+ <position ctrllimited="true" ctrlrange="0.021 0.057" joint="vx300s_right/left_finger" kp="200" user="1"/>
+ <position ctrllimited="true" ctrlrange="-0.057 -0.021" joint="vx300s_right/right_finger" kp="200" user="1"/>
+
+ </actuator>
+
+ <keyframe>
+ <key qpos="0 -0.96 1.16 0 -0.3 0 0.024 -0.024 0 -0.96 1.16 0 -0.3 0 0.024 -0.024 0.2 0.5 0.05 1 0 0 0"/>
+ </keyframe>
+
+
+</mujoco>
\ No newline at end of file
diff --git a/constants.py b/constants.py
index f445350..59afe22 100644
--- a/constants.py
+++ b/constants.py
@@ -1,7 +1,8 @@
+import os
import pathlib
### Task parameters
-DATA_DIR = '<put your data dir here>'
+DATA_DIR = os.environ.get('ACT_DATA_DIR', str(pathlib.Path(__file__).parent.parent.resolve() / 'data'))
SIM_TASK_CONFIGS = {
'sim_transfer_cube_scripted':{
'dataset_dir': DATA_DIR + '/sim_transfer_cube_scripted',
diff --git a/detr/main.py b/detr/main.py
index 3c4a339..1a5592b 100644
--- a/detr/main.py
+++ b/detr/main.py
@@ -1,5 +1,6 @@
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
import argparse
+import os
from pathlib import Path
import numpy as np
@@ -63,6 +64,11 @@ def get_args_parser():
parser.add_argument('--kl_weight', action='store', type=int, help='KL Weight', required=False)
parser.add_argument('--chunk_size', action='store', type=int, help='chunk_size', required=False)
parser.add_argument('--temporal_agg', action='store_true')
+ parser.add_argument('--device', action='store', type=str, default=os.environ.get('ACT_DEVICE', 'mps'))
+ parser.add_argument('--num_rollouts', action='store', type=int)
+ parser.add_argument('--eval_seed', action='store', type=int)
+ parser.add_argument('--eval_ckpt_names', action='store', type=str, nargs='+')
+ parser.add_argument('--no_save_video', action='store_true')
return parser
@@ -75,7 +81,7 @@ def build_ACT_model_and_optimizer(args_override):
setattr(args, k, v)
model = build_ACT_model(args)
- model.cuda()
+ model.to(args.device)
param_dicts = [
{"params": [p for n, p in model.named_parameters() if "backbone" not in n and p.requires_grad]},
@@ -98,7 +104,7 @@ def build_CNNMLP_model_and_optimizer(args_override):
setattr(args, k, v)
model = build_CNNMLP_model(args)
- model.cuda()
+ model.to(args.device)
param_dicts = [
{"params": [p for n, p in model.named_parameters() if "backbone" not in n and p.requires_grad]},
diff --git a/imitate_episodes.py b/imitate_episodes.py
index 34f9a37..21d7e3a 100644
--- a/imitate_episodes.py
+++ b/imitate_episodes.py
@@ -17,6 +17,7 @@ from policy import ACTPolicy, CNNMLPPolicy
from visualize_episodes import save_videos
from sim_env import BOX_POSE
+from cfnbc.act_hooks import make_train_logger, make_rollout_logger, enabled as rerun_enabled # no-op unless ACT_RERUN=1
import IPython
e = IPython.embed
@@ -32,6 +33,7 @@ def main(args):
batch_size_train = args['batch_size']
batch_size_val = args['batch_size']
num_epochs = args['num_epochs']
+ device = torch.device(args['device'])
# get task parameters
is_sim = task_name[:4] == 'sim_'
@@ -85,14 +87,18 @@ def main(args):
'seed': args['seed'],
'temporal_agg': args['temporal_agg'],
'camera_names': camera_names,
- 'real_robot': not is_sim
+ 'real_robot': not is_sim,
+ 'device': device,
+ 'num_rollouts': args['num_rollouts'],
+ 'eval_seed': args['eval_seed'],
+ 'save_video': not args['no_save_video'],
}
if is_eval:
- ckpt_names = [f'policy_best.ckpt']
+ ckpt_names = args['eval_ckpt_names']
results = []
for ckpt_name in ckpt_names:
- success_rate, avg_return = eval_bc(config, ckpt_name, save_episode=True)
+ success_rate, avg_return = eval_bc(config, ckpt_name, save_episode=config['save_video'])
results.append([ckpt_name, success_rate, avg_return])
for ckpt_name, success_rate, avg_return in results:
@@ -138,18 +144,19 @@ def make_optimizer(policy_class, policy):
return optimizer
-def get_image(ts, camera_names):
+def get_image(ts, camera_names, device):
curr_images = []
for cam_name in camera_names:
curr_image = rearrange(ts.observation['images'][cam_name], 'h w c -> c h w')
curr_images.append(curr_image)
curr_image = np.stack(curr_images, axis=0)
- curr_image = torch.from_numpy(curr_image / 255.0).float().cuda().unsqueeze(0)
+ curr_image = torch.from_numpy(curr_image / 255.0).float().to(device).unsqueeze(0)
return curr_image
def eval_bc(config, ckpt_name, save_episode=True):
- set_seed(1000)
+ set_seed(config['eval_seed'])
+ device = config['device']
ckpt_dir = config['ckpt_dir']
state_dim = config['state_dim']
real_robot = config['real_robot']
@@ -165,9 +172,9 @@ def eval_bc(config, ckpt_name, save_episode=True):
# load policy and stats
ckpt_path = os.path.join(ckpt_dir, ckpt_name)
policy = make_policy(policy_class, policy_config)
- loading_status = policy.load_state_dict(torch.load(ckpt_path))
+ loading_status = policy.load_state_dict(torch.load(ckpt_path, map_location=device))
print(loading_status)
- policy.cuda()
+ policy.to(device)
policy.eval()
print(f'Loaded: {ckpt_path}')
stats_path = os.path.join(ckpt_dir, f'dataset_stats.pkl')
@@ -195,18 +202,22 @@ def eval_bc(config, ckpt_name, save_episode=True):
max_timesteps = int(max_timesteps * 1) # may increase for real-world tasks
- num_rollouts = 50
+ num_rollouts = config['num_rollouts']
+ # draw all initial object poses up front from the eval seed (same sequence as drawing them per rollout)
+ if 'sim_transfer_cube' in task_name:
+ init_poses = [sample_box_pose() for _ in range(num_rollouts)]
+ elif 'sim_insertion' in task_name:
+ init_poses = [np.concatenate(sample_insertion_pose()) for _ in range(num_rollouts)]
+ rlog = make_rollout_logger(config, ckpt_name)
episode_returns = []
highest_rewards = []
for rollout_id in range(num_rollouts):
rollout_id += 0
### set task
- if 'sim_transfer_cube' in task_name:
- BOX_POSE[0] = sample_box_pose() # used in sim reset
- elif 'sim_insertion' in task_name:
- BOX_POSE[0] = np.concatenate(sample_insertion_pose()) # used in sim reset
+ BOX_POSE[0] = init_poses[rollout_id] # used in sim reset
ts = env.reset()
+ rlog.start(rollout_id, init_poses[rollout_id])
### onscreen render
if onscreen_render:
@@ -216,9 +227,9 @@ def eval_bc(config, ckpt_name, save_episode=True):
### evaluation loop
if temporal_agg:
- all_time_actions = torch.zeros([max_timesteps, max_timesteps+num_queries, state_dim]).cuda()
+ all_time_actions = torch.zeros([max_timesteps, max_timesteps+num_queries, state_dim]).to(device)
- qpos_history = torch.zeros((1, max_timesteps, state_dim)).cuda()
+ qpos_history = torch.zeros((1, max_timesteps, state_dim)).to(device)
image_list = [] # for visualization
qpos_list = []
target_qpos_list = []
@@ -239,14 +250,17 @@ def eval_bc(config, ckpt_name, save_episode=True):
image_list.append({'main': obs['image']})
qpos_numpy = np.array(obs['qpos'])
qpos = pre_process(qpos_numpy)
- qpos = torch.from_numpy(qpos).float().cuda().unsqueeze(0)
+ qpos = torch.from_numpy(qpos).float().to(device).unsqueeze(0)
qpos_history[:, t] = qpos
- curr_image = get_image(ts, camera_names)
+ curr_image = get_image(ts, camera_names, device)
### query policy
if config['policy_class'] == "ACT":
if t % query_frequency == 0:
all_actions = policy(qpos, curr_image)
+ chunk = post_process(all_actions[0].cpu().numpy()) if rerun_enabled() else None
+ else:
+ chunk = None
if temporal_agg:
all_time_actions[[t], t:t+num_queries] = all_actions
actions_for_curr_step = all_time_actions[:, t]
@@ -255,7 +269,7 @@ def eval_bc(config, ckpt_name, save_episode=True):
k = 0.01
exp_weights = np.exp(-k * np.arange(len(actions_for_curr_step)))
exp_weights = exp_weights / exp_weights.sum()
- exp_weights = torch.from_numpy(exp_weights).cuda().unsqueeze(dim=1)
+ exp_weights = torch.from_numpy(exp_weights).float().to(device).unsqueeze(dim=1) # float32: MPS has no float64
raw_action = (actions_for_curr_step * exp_weights).sum(dim=0, keepdim=True)
else:
raw_action = all_actions[:, t % query_frequency]
@@ -270,7 +284,9 @@ def eval_bc(config, ckpt_name, save_episode=True):
target_qpos = action
### step the environment
+ obs_t = ts.observation
ts = env.step(target_qpos)
+ rlog.step(t, obs_t, env.physics, target_qpos, chunk if config['policy_class'] == 'ACT' else None, ts.reward)
### for visualization
qpos_list.append(qpos_numpy)
@@ -287,6 +303,7 @@ def eval_bc(config, ckpt_name, save_episode=True):
episode_returns.append(episode_return)
episode_highest_reward = np.max(rewards)
highest_rewards.append(episode_highest_reward)
+ rlog.end(episode_highest_reward, episode_highest_reward == env_max_reward)
print(f'Rollout {rollout_id}\n{episode_return=}, {episode_highest_reward=}, {env_max_reward=}, Success: {episode_highest_reward==env_max_reward}')
if save_episode:
@@ -303,7 +320,15 @@ def eval_bc(config, ckpt_name, save_episode=True):
print(summary_str)
# save success rate to txt
- result_file_name = 'result_' + ckpt_name.split('.')[0] + '.txt'
+ result_tag = ckpt_name.split('.')[0] + ('_tagg' if temporal_agg else '_notagg') + f'_seed{config["eval_seed"]}_n{num_rollouts}'
+ if os.environ.get('ACT_NUISANCE'):
+ result_tag += '_' + os.environ['ACT_NUISANCE']
+ with open(os.path.join(ckpt_dir, f'rollouts_{result_tag}.csv'), 'w') as f:
+ f.write('rollout_id,init_pose,episode_return,highest_reward,success\n')
+ for i in range(num_rollouts):
+ pose_str = ' '.join(f'{x:.8f}' for x in init_poses[i])
+ f.write(f'{i},{pose_str},{episode_returns[i]},{highest_rewards[i]},{int(highest_rewards[i] == env_max_reward)}\n')
+ result_file_name = 'result_' + result_tag + '.txt'
with open(os.path.join(ckpt_dir, result_file_name), 'w') as f:
f.write(summary_str)
f.write(repr(episode_returns))
@@ -313,9 +338,9 @@ def eval_bc(config, ckpt_name, save_episode=True):
return success_rate, avg_return
-def forward_pass(data, policy):
+def forward_pass(data, policy, device):
image_data, qpos_data, action_data, is_pad = data
- image_data, qpos_data, action_data, is_pad = image_data.cuda(), qpos_data.cuda(), action_data.cuda(), is_pad.cuda()
+ image_data, qpos_data, action_data, is_pad = image_data.to(device), qpos_data.to(device), action_data.to(device), is_pad.to(device)
return policy(qpos_data, image_data, action_data, is_pad) # TODO remove None
@@ -325,12 +350,14 @@ def train_bc(train_dataloader, val_dataloader, config):
seed = config['seed']
policy_class = config['policy_class']
policy_config = config['policy_config']
+ device = config['device']
set_seed(seed)
policy = make_policy(policy_class, policy_config)
- policy.cuda()
+ policy.to(device)
optimizer = make_optimizer(policy_class, policy)
+ tlog = make_train_logger(config)
train_history = []
validation_history = []
@@ -343,10 +370,11 @@ def train_bc(train_dataloader, val_dataloader, config):
policy.eval()
epoch_dicts = []
for batch_idx, data in enumerate(val_dataloader):
- forward_dict = forward_pass(data, policy)
+ forward_dict = forward_pass(data, policy, device)
epoch_dicts.append(forward_dict)
epoch_summary = compute_dict_mean(epoch_dicts)
validation_history.append(epoch_summary)
+ tlog.log_epoch(epoch, 'val', epoch_summary)
epoch_val_loss = epoch_summary['loss']
if epoch_val_loss < min_val_loss:
@@ -362,15 +390,17 @@ def train_bc(train_dataloader, val_dataloader, config):
policy.train()
optimizer.zero_grad()
for batch_idx, data in enumerate(train_dataloader):
- forward_dict = forward_pass(data, policy)
+ forward_dict = forward_pass(data, policy, device)
# backward
loss = forward_dict['loss']
loss.backward()
optimizer.step()
optimizer.zero_grad()
train_history.append(detach_dict(forward_dict))
+ tlog.log_batch(train_history[-1])
epoch_summary = compute_dict_mean(train_history[(batch_idx+1)*epoch:(batch_idx+1)*(epoch+1)])
epoch_train_loss = epoch_summary['loss']
+ tlog.log_epoch(epoch, 'train', epoch_summary)
print(f'Train loss: {epoch_train_loss:.5f}')
summary_string = ''
for k, v in epoch_summary.items():
@@ -389,6 +419,8 @@ def train_bc(train_dataloader, val_dataloader, config):
ckpt_path = os.path.join(ckpt_dir, f'policy_epoch_{best_epoch}_seed_{seed}.ckpt')
torch.save(best_state_dict, ckpt_path)
print(f'Training finished:\nSeed {seed}, val loss {min_val_loss:.6f} at epoch {best_epoch}')
+ tlog.log_best(best_epoch, float(min_val_loss))
+ tlog.close()
# save training curves
plot_history(train_history, validation_history, num_epochs, ckpt_dir, seed)
@@ -431,5 +463,12 @@ if __name__ == '__main__':
parser.add_argument('--hidden_dim', action='store', type=int, help='hidden_dim', required=False)
parser.add_argument('--dim_feedforward', action='store', type=int, help='dim_feedforward', required=False)
parser.add_argument('--temporal_agg', action='store_true')
-
+
+ # device / eval options (cfnbc baseline)
+ parser.add_argument('--device', action='store', type=str, default=os.environ.get('ACT_DEVICE', 'mps'), help='torch device (env ACT_DEVICE)')
+ parser.add_argument('--num_rollouts', action='store', type=int, default=100)
+ parser.add_argument('--eval_seed', action='store', type=int, default=1000)
+ parser.add_argument('--eval_ckpt_names', action='store', type=str, nargs='+', default=['policy_last.ckpt'])
+ parser.add_argument('--no_save_video', action='store_true')
+
main(vars(parser.parse_args()))
diff --git a/record_sim_episodes.py b/record_sim_episodes.py
index 253fdea..695d1bb 100644
--- a/record_sim_episodes.py
+++ b/record_sim_episodes.py
@@ -4,6 +4,7 @@ import numpy as np
import argparse
import matplotlib.pyplot as plt
import h5py
+import mujoco
from constants import PUPPET_GRIPPER_POSITION_NORMALIZE_FN, SIM_TASK_CONFIGS
from ee_sim_env import make_ee_sim_env
@@ -27,6 +28,8 @@ def main(args):
dataset_dir = args['dataset_dir']
num_episodes = args['num_episodes']
onscreen_render = args['onscreen_render']
+ keep_failed = args['keep_failed']
+ np.random.seed(args['seed'])
inject_noise = False
render_cam_name = 'angle'
@@ -43,8 +46,12 @@ def main(args):
raise NotImplementedError
success = []
- for episode_idx in range(num_episodes):
- print(f'{episode_idx=}')
+ attempt_log = [] # one row per attempt, including regenerated failures
+ episode_idx = 0
+ attempt_idx = -1
+ while episode_idx < num_episodes:
+ attempt_idx += 1
+ print(f'{episode_idx=} {attempt_idx=}')
print('Rollout out EE space scripted policy')
# setup the environment
env = make_ee_sim_env(task_name)
@@ -67,6 +74,7 @@ def main(args):
episode_return = np.sum([ts.reward for ts in episode[1:]])
episode_max_reward = np.max([ts.reward for ts in episode[1:]])
+ ee_success = int(episode_max_reward == env.task.max_reward)
if episode_max_reward == env.task.max_reward:
print(f"{episode_idx=} Successful, {episode_return=}")
else:
@@ -95,6 +103,7 @@ def main(args):
ts = env.reset()
episode_replay = [ts]
+ sim_states = [get_sim_state(env.physics)] # full physics state aligned with episode_replay
# setup plotting
if onscreen_render:
ax = plt.subplot()
@@ -104,21 +113,29 @@ def main(args):
action = joint_traj[t]
ts = env.step(action)
episode_replay.append(ts)
+ sim_states.append(get_sim_state(env.physics))
if onscreen_render:
plt_img.set_data(ts.observation['images'][render_cam_name])
plt.pause(0.02)
episode_return = np.sum([ts.reward for ts in episode_replay[1:]])
episode_max_reward = np.max([ts.reward for ts in episode_replay[1:]])
+ replay_success = int(episode_max_reward == env.task.max_reward)
if episode_max_reward == env.task.max_reward:
- success.append(1)
print(f"{episode_idx=} Successful, {episode_return=}")
else:
- success.append(0)
print(f"{episode_idx=} Failed")
plt.close()
+ saved = bool(replay_success or keep_failed)
+ attempt_log.append((attempt_idx, episode_idx if saved else -1, ' '.join(f'{x:.8f}' for x in subtask_info),
+ ee_success, replay_success, episode_max_reward, episode_return))
+ if not saved:
+ print(f'Discarding failed attempt {attempt_idx}, regenerating episode {episode_idx}\n')
+ continue
+ success.append(replay_success)
+
"""
For each timestep:
observations
@@ -146,6 +163,7 @@ def main(args):
# len(joint_traj) i.e. actions: max_timesteps
# len(episode_replay) i.e. time steps: max_timesteps + 1
max_timesteps = len(joint_traj)
+ sim_states = sim_states[:max_timesteps] # same timesteps as the observations saved below
while joint_traj:
action = joint_traj.pop(0)
ts = episode_replay.pop(0)
@@ -173,10 +191,44 @@ def main(args):
for name, array in data_dict.items():
root[name][...] = array
+
+ # extra (cfnbc): full MuJoCo state per timestep, aligned with /observations, for exact re-rendering
+ root.attrs['mujoco_version'] = mujoco.__version__
+ root.attrs['record_seed'] = args['seed']
+ root.attrs['attempt_idx'] = attempt_idx
+ root.attrs['success'] = replay_success
+ root.attrs['init_env_state'] = subtask_info
+ sim = root.create_group('sim_state')
+ for key in sim_states[0]:
+ sim.create_dataset(key, data=np.stack([st[key] for st in sim_states]))
print(f'Saving: {time.time() - t0:.1f} secs\n')
+ episode_idx += 1
print(f'Saved to {dataset_dir}')
- print(f'Success: {np.sum(success)} / {len(success)}')
+ print(f'Success: {np.sum(success)} / {len(success)} saved, {len(attempt_log)} attempts')
+
+ log_path = args['success_log'] or os.path.join(dataset_dir, 'demo_success.csv')
+ os.makedirs(os.path.dirname(os.path.abspath(log_path)), exist_ok=True)
+ with open(log_path, 'w') as f:
+ f.write('attempt_idx,episode_idx,init_env_state,ee_success,replay_success,max_reward,episode_return\n')
+ for row in attempt_log:
+ f.write(','.join(str(x) for x in row) + '\n')
+ print(f'Success log: {log_path}')
+
+
+def get_sim_state(physics):
+ """Full physics state; set qpos/qvel/act (+ctrl, mocap) and call mj_forward to restore for rendering."""
+ d = physics.data
+ return {
+ 'time': np.array(d.time),
+ 'qpos': d.qpos.copy(), # includes red_box free joint: qpos[16:23] = xyz + quat (wxyz)
+ 'qvel': d.qvel.copy(),
+ 'act': d.act.copy(),
+ 'ctrl': d.ctrl.copy(),
+ 'mocap_pos': d.mocap_pos.copy(),
+ 'mocap_quat': d.mocap_quat.copy(),
+ 'physics_state': physics.get_state().copy(), # dm_control flat state (qpos, qvel, act)
+ }
if __name__ == '__main__':
parser = argparse.ArgumentParser()
@@ -184,6 +236,9 @@ if __name__ == '__main__':
parser.add_argument('--dataset_dir', action='store', type=str, help='dataset saving dir', required=True)
parser.add_argument('--num_episodes', action='store', type=int, help='num_episodes', required=False)
parser.add_argument('--onscreen_render', action='store_true')
+ parser.add_argument('--seed', action='store', type=int, default=0, help='numpy seed for box poses')
+ parser.add_argument('--keep_failed', action='store_true', help='save failed episodes instead of regenerating')
+ parser.add_argument('--success_log', action='store', type=str, default=None, help='csv path (default <dataset_dir>/demo_success.csv)')
main(vars(parser.parse_args()))
diff --git a/sim_env.py b/sim_env.py
index b79b935..4a5abe8 100644
--- a/sim_env.py
+++ b/sim_env.py
@@ -17,6 +17,10 @@ e = IPython.embed
BOX_POSE = [None] # to be changed from outside
+# cfnbc: ACT_NUISANCE=<condition> loads the nuisance scene and applies that condition at every reset;
+# ACT_RENDER_CAMS=top,... restricts rendered cameras (default: all three, as upstream).
+RENDER_CAMS = {'top': 'top', 'angle': 'angle', 'vis': 'front_close'}
+
def make_sim_env(task_name):
"""
Environment for simulated robot bi-manual manipulation, with joint position control
@@ -36,9 +40,12 @@ def make_sim_env(task_name):
"images": {"main": (480x640x3)} # h, w, c, dtype='uint8'
"""
if 'sim_transfer_cube' in task_name:
- xml_path = os.path.join(XML_DIR, f'bimanual_viperx_transfer_cube.xml')
+ from cfnbc import nuisance
+ nuisance_name = nuisance.from_env()
+ xml_file = nuisance.NUISANCE_XML if nuisance_name else 'bimanual_viperx_transfer_cube.xml'
+ xml_path = os.path.join(XML_DIR, xml_file)
physics = mujoco.Physics.from_xml_path(xml_path)
- task = TransferCubeTask(random=False)
+ task = TransferCubeTask(random=False, nuisance_name=nuisance_name)
env = control.Environment(physics, task, time_limit=20, control_timestep=DT,
n_sub_steps=None, flat_observation=False)
elif 'sim_insertion' in task_name:
@@ -107,9 +114,9 @@ class BimanualViperXTask(base.Task):
obs['qvel'] = self.get_qvel(physics)
obs['env_state'] = self.get_env_state(physics)
obs['images'] = dict()
- obs['images']['top'] = physics.render(height=480, width=640, camera_id='top')
- obs['images']['angle'] = physics.render(height=480, width=640, camera_id='angle')
- obs['images']['vis'] = physics.render(height=480, width=640, camera_id='front_close')
+ cams = [c for c in os.environ.get('ACT_RENDER_CAMS', 'top,angle,vis').split(',') if c]
+ for cam in cams:
+ obs['images'][cam] = physics.render(height=480, width=640, camera_id=RENDER_CAMS[cam])
return obs
@@ -119,9 +126,10 @@ class BimanualViperXTask(base.Task):
class TransferCubeTask(BimanualViperXTask):
- def __init__(self, random=None):
+ def __init__(self, random=None, nuisance_name=None):
super().__init__(random=random)
self.max_reward = 4
+ self.nuisance_name = nuisance_name
def initialize_episode(self, physics):
"""Sets the state of the environment at the start of each episode."""
@@ -134,6 +142,9 @@ class TransferCubeTask(BimanualViperXTask):
physics.named.data.qpos[-7:] = BOX_POSE[0]
# print(f"{BOX_POSE=}")
super().initialize_episode(physics)
+ if self.nuisance_name:
+ from cfnbc import nuisance
+ nuisance.apply(physics, self.nuisance_name, BOX_POSE[0])
@staticmethod
def get_env_state(physics):
diff --git a/utils.py b/utils.py
index d90b782..fa6ef8d 100644
--- a/utils.py
+++ b/utils.py
@@ -122,8 +122,11 @@ def load_data(dataset_dir, num_episodes, camera_names, batch_size_train, batch_s
# construct dataset and dataloader
train_dataset = EpisodicDataset(train_indices, dataset_dir, camera_names, norm_stats)
val_dataset = EpisodicDataset(val_indices, dataset_dir, camera_names, norm_stats)
- train_dataloader = DataLoader(train_dataset, batch_size=batch_size_train, shuffle=True, pin_memory=True, num_workers=1, prefetch_factor=1)
- val_dataloader = DataLoader(val_dataset, batch_size=batch_size_val, shuffle=True, pin_memory=True, num_workers=1, prefetch_factor=1)
+ # cfnbc: ACT_NUM_WORKERS overrides upstream num_workers=1 (0 = load in main process, seeded numpy RNG)
+ num_workers = int(os.environ.get('ACT_NUM_WORKERS', 1))
+ loader_kwargs = dict(num_workers=num_workers, prefetch_factor=1) if num_workers > 0 else dict(num_workers=0)
+ train_dataloader = DataLoader(train_dataset, batch_size=batch_size_train, shuffle=True, pin_memory=True, **loader_kwargs)
+ val_dataloader = DataLoader(val_dataset, batch_size=batch_size_val, shuffle=True, pin_memory=True, **loader_kwargs)
return train_dataloader, val_dataloader, norm_stats, train_dataset.is_sim