-
Notifications
You must be signed in to change notification settings - Fork 1
/
demos.py
282 lines (241 loc) · 8.23 KB
/
demos.py
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
"""Data collection script."""
import os
import pickle
import random
import numpy as np
import wandb
import tasks
from environments.environment import Environment
import ipdb
st = ipdb.set_trace
def main(cfg, env):
# Set task
task = tasks.names[cfg['task']]()
task.mode = cfg['mode']
task.name = cfg['task']
env.set_task(task)
# st()
record = cfg['record']['save_video']
save_data = cfg['save_data'] # if False, just loop for debugging
force_generate = cfg['force_generate']
# Initialize scripted oracle agent
agent = task.oracle(env)
data_path = os.path.join(
cfg['data_dir'], "{}-{}".format(cfg['task'], task.mode)
)
print(f"Saving to: {data_path}")
os.makedirs(data_path, exist_ok=True)
# Train seeds are even and val/test seeds are odd.
# Test seeds are offset by 10000
seed, n_episodes = _get_seed_and_episodes(data_path, force_generate)
if seed < 0 or not save_data:
seed = _get_init_seed(task.mode)
n_episodes = 0
# Collect training data from oracle demonstrations.
lang_goals_video = []
while n_episodes < cfg['n_demos']:
episode, total_reward = [], 0
seed += 2
# Set seeds.
np.random.seed(seed)
random.seed(seed)
print('Oracle demo: {}/{} | Seed: {}'.format(
n_episodes + 1, cfg['n_demos'], seed))
env.set_task(task)
obs = env.reset()
if env.failed_datagen:
env.set_failed_dategen(False)
continue
info = env.info
reward = 0
lang_goals_video.append(info['lang_goal'])
# Start video recording (NOTE: super slow)
if record:
env.start_rec(f'{n_episodes+1:06d}')
# Rollout expert policy
for i in range(task.max_steps):
# target reward for a successful action
goal_reward = task.goals[0][-1]
expected_steps = len(task.goals[0][0])
target_reward = goal_reward / expected_steps - 0.003
done_multitask = (i == task.max_steps - 1)
# act
act = agent.act(obs, info)
if act is None:
break
episode.append((obs, act, reward, info))
lang_goal = info['lang_goal']
# take a step
obs, _, _, _ = env.step(act)
_, _, obj_mask = task.get_true_image(env)
reward = env.task.reward(
oracle=True, datagen=True, done_multitask=done_multitask,
obj_mask=obj_mask)
done = env.task.done()
info = env.info
# did we hit the target reward?
if reward < target_reward and cfg['discard_imperfect']:
print(f"discraded: {reward} is not max: {target_reward}")
break
# we hit, so update reward
total_reward += reward
print(
f'Total Reward: {total_reward:.3f}'
f' | Done: {done} | Goal: {lang_goal}'
)
if done:
break
episode.append((obs, None, reward, info))
# End video recording
if record:
env.end_rec()
# Only save completed demonstrations.
if save_data and total_reward > 0.99:
_store_demo(data_path, seed, episode, n_episodes)
n_episodes += 1
elif total_reward <= 0.99:
print("discarded")
lang_goals_video = lang_goals_video[:-1]
else: # reward is ok but don't save
n_episodes += 1
# Play videos in wandb
if record:
_show_videos(cfg, lang_goals_video)
def _init_env(cfg):
return Environment(
cfg['assets_root'],
disp=cfg['disp'],
shared_memory=cfg['shared_memory'],
hz=480,
record_cfg=cfg['record'],
debug=False,
constant_bg=True
)
def _get_seed_and_episodes(data_path, force_generate=False):
color_path = os.path.join(data_path, 'action')
n_episodes = 0
max_seed = -1
if os.path.exists(color_path) and not force_generate:
for fname in sorted(os.listdir(color_path)):
if '.pkl' in fname:
seed = int(fname[(fname.find('-') + 1):-4])
n_episodes += 1
max_seed = max(max_seed, seed)
else:
print(f"{color_path} doesn't exist")
return max_seed, n_episodes
def _get_init_seed(mode):
if mode == 'train':
seed = -2
elif mode == 'val':
seed = -1
elif mode == 'test':
seed = -1 + 10000
else:
raise Exception("Invalid mode. Valid options: train, val, test")
return seed
def _dump(data, field, data_path, seed, n_episodes):
field_path = os.path.join(data_path, field)
os.makedirs(field_path, exist_ok=True)
fname = f'{n_episodes:06d}-{seed}.pkl' # -{len(episode):06d}
with open(os.path.join(field_path, fname), 'wb') as f:
pickle.dump(data, f)
def _store_demo(data_path, seed, episode, n_episodes):
color, depth, action, reward, info = [], [], [], [], []
for obs, act, r, i in episode:
color.append(obs['color'])
depth.append(obs['depth'])
action.append(act)
reward.append(r)
info.append(i)
color = np.uint8(color)
depth = np.float32(depth)
_dump(color, 'color', data_path, seed, n_episodes)
_dump(depth, 'depth', data_path, seed, n_episodes)
_dump(action, 'action', data_path, seed, n_episodes)
_dump(reward, 'reward', data_path, seed, n_episodes)
_dump(info, 'info', data_path, seed, n_episodes)
def _show_videos(cfg, lang_goals_video):
folder = cfg['record']['save_video_path']
for subdir, _, files in os.walk(folder):
for i, file_ in enumerate(sorted(files)):
if file_.endswith('mp4'):
print(file_)
video = wandb.Video(
os.path.join(folder, subdir, file_),
fps=25,
format="mp4",
caption=lang_goals_video[i]
)
wandb.log({"vis": video})
if __name__ == '__main__':
cfg = {
'assets_root': 'environments/assets/',
'data_dir': 'benchmark_data/',
'discard_imperfect': True,
'disp': False,
'record': {
'save_video': False,
'save_video_path':
'benchmark_data/',
'add_text': False,
'fps': 25,
'video_height': 640,
'video_width': 720,
},
'save_data': True,
'shared_memory': False,
'force_generate': False ,
}
env = _init_env(cfg)
task_list = [
# cliport tasks
'assembling-kits-seq-seen-colors',
'packing-seen-google-objects-group',
'packing-seen-google-objects-seq',
'put-block-in-bowl-seen-colors',
'assembling-kits-seq-unseen-colors',
'packing-unseen-google-objects-group',
'packing-unseen-google-objects-seq',
'put-block-in-bowl-unseen-colors',
# spatial relations
'right-seen-colors',
'above-seen-colors',
'below-seen-colors',
'left-seen-colors',
'right-unseen-colors',
'above-unseen-colors',
'below-unseen-colors',
'left-unseen-colors',
#composition relations
'composition-seen-colors',
'composition-unseen-colors',
'composition-seen-colors-group',
'composition-unseen-colors-group'
# shape
'circle-seen-colors',
'line-seen-colors',
'circle-unseen-colors',
'line-unseen-colors'
]
splits = ['train', 'val', 'test']
# splits = ['test']
if cfg['record']['save_video']:
splits = ['train']
cfg['save_data'] = False
wandb.init(project="robot", name="data_vis")
for name in task_list:
for split in splits:
if split == "val":
n_demos = 50
elif split == "test":
n_demos = 100
else:
n_demos = 100
n_demos = 50 if split == 'val' else 100
if cfg['record']['save_video']:
n_demos = 2
cfg['task'] = name
cfg['mode'] = split
cfg['n_demos'] = n_demos
main(cfg, env)