-
Notifications
You must be signed in to change notification settings - Fork 9
/
Copy pathtrain_low_level.py
194 lines (173 loc) · 6.08 KB
/
train_low_level.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
"""Main script."""
import os
import hydra
import torch
from rk_diffuser.dataset.rl_bench_dataset import RLBenchDataset
from rk_diffuser.models.multi_level_diffusion import MultiLevelDiffusion
from rk_diffuser.robot import DiffRobot
from rk_diffuser.trainer import Trainer
from rk_diffuser.utils import load_checkpoint
VALID_DIFFUSION_VARS = ["gripper_poses", "joint_positions", "multi"]
def _create_agent_fn(
cfgs,
device,
diffusion_var="multi",
sim=True,
diff_optim=True,
diff_optim_steps=100,
diff_lr=10,
pose_augment=False,
):
robot = None
file_path = os.path.join(*__file__.split("/")[:-1])
if diffusion_var in ["joint_positions", "multi"]:
robot = DiffRobot(
os.path.join("/", file_path, "panda_urdf/panda.urdf"),
)
if robot is not None:
robot.to(device)
assert diffusion_var in VALID_DIFFUSION_VARS
if diffusion_var == "multi":
diffusion_pose = hydra.utils.instantiate(
cfgs,
diffusion_var="gripper_poses",
)
diffusion_joints = hydra.utils.instantiate(
cfgs,
diffusion_var="joint_positions",
)
diffusion = MultiLevelDiffusion(
{
"gripper_poses": diffusion_pose,
"joint_positions": diffusion_joints,
},
diff_optim=diff_optim,
diff_optim_steps=diff_optim_steps,
diff_lr=diff_lr,
pose_augment=pose_augment,
sim=sim,
)
else:
diffusion = hydra.utils.instantiate(
cfgs,
diffusion_var=diffusion_var,
)
diffusion.to(device)
return robot, diffusion
@hydra.main(
config_path="cfgs",
config_name="diffuser_config",
version_base=None,
)
def main(cfgs):
device = "cuda" if torch.cuda.is_available() else "cpu"
robot, diffusion = _create_agent_fn(
cfgs.method,
device,
diffusion_var=cfgs.diffusion_var,
sim=cfgs.env.name == "sim",
diff_optim=cfgs.diff_optim,
diff_optim_steps=cfgs.diff_optim_steps,
diff_lr=cfgs.diff_lr,
pose_augment=False,
)
if os.path.exists(cfgs.load_model_path):
load_checkpoint(diffusion, cfgs.load_model_path, cfgs.method.backbone)
diffusion = diffusion.to(device)
assert os.path.isdir(cfgs.env.data_path)
if cfgs.env.name == "sim":
dataset = RLBenchDataset(
cfgs.env.tasks,
cfgs.env.tasks_ratio,
cfgs.env.cameras,
cfgs.env.num_episodes,
data_raw_path=os.path.join(cfgs.env.data_path, "train"),
traj_len=cfgs.method.horizon,
frame_skips=cfgs.frame_skips,
observation_dim=cfgs.method.observation_dim,
rank_bins=cfgs.method.rank_bins,
robot=robot,
diffusion_var=cfgs.diffusion_var,
demo_aug_ratio=cfgs.env.demo_aug_ratio,
demo_aug_min_len=cfgs.env.demo_aug_min_len,
use_cached=cfgs.use_cached,
ds_img_size=cfgs.ds_img_size,
)
eval_dataset = RLBenchDataset(
cfgs.env.tasks,
cfgs.env.tasks_ratio,
cfgs.env.cameras,
cfgs.env.num_episodes // 2,
data_raw_path=os.path.join(cfgs.env.data_path, "eval"),
traj_len=cfgs.method.horizon,
frame_skips=cfgs.frame_skips,
observation_dim=cfgs.method.observation_dim,
rank_bins=cfgs.method.rank_bins,
robot=robot,
diffusion_var=cfgs.diffusion_var,
training=False,
demo_aug_ratio=cfgs.env.demo_aug_ratio,
demo_aug_min_len=cfgs.env.demo_aug_min_len,
use_cached=cfgs.use_cached,
ds_img_size=cfgs.ds_img_size,
)
else:
# RLBench complains when importing cv2 before it so we move it here
from rk_diffuser.dataset.realworld_dataset import RealWorldDataset
dataset = RealWorldDataset(
cfgs.env.tasks,
cfgs.env.cameras,
cfgs.env.num_episodes,
data_raw_path=cfgs.env.data_path,
traj_len=cfgs.method.horizon,
frame_skips=cfgs.frame_skips,
observation_dim=cfgs.method.observation_dim,
rank_bins=cfgs.method.rank_bins,
robot=robot,
diffusion_var=cfgs.diffusion_var,
demo_aug_ratio=cfgs.env.demo_aug_ratio,
demo_aug_min_len=cfgs.env.demo_aug_min_len,
camera_extrinsics=cfgs.env.camera_extrinsics,
load_processed_data=cfgs.env.load_processed_data,
save_processed_data=cfgs.env.save_processed_data,
)
eval_dataset = RealWorldDataset(
cfgs.env.tasks,
cfgs.env.cameras,
cfgs.env.num_episodes_eval,
data_raw_path=cfgs.env.data_path,
traj_len=cfgs.method.horizon,
frame_skips=cfgs.frame_skips,
observation_dim=cfgs.method.observation_dim,
rank_bins=cfgs.method.rank_bins,
robot=robot,
diffusion_var=cfgs.diffusion_var,
demo_aug_ratio=cfgs.env.demo_aug_ratio,
demo_aug_min_len=cfgs.env.demo_aug_min_len,
camera_extrinsics=cfgs.env.camera_extrinsics,
training=False,
load_processed_data=cfgs.env.load_processed_data,
save_processed_data=cfgs.env.save_processed_data,
)
trainer = Trainer(
cfgs=cfgs,
diffusion_model=diffusion,
dataset=dataset,
eval_dataset=eval_dataset,
train_batch_size=cfgs.batch_size,
log=cfgs.log,
log_freq=cfgs.log_freq,
save_freq=cfgs.save_freq,
scene_bounds=cfgs.env.scene_bounds,
project_name=cfgs.project_name,
online_eval=cfgs.online_eval,
headless=cfgs.headless,
rank_bins=cfgs.method.rank_bins,
robot=robot,
diffusion_var=cfgs.diffusion_var,
online_eval_start=cfgs.online_eval_start,
action_mode=cfgs.action_mode,
)
trainer.train(cfgs.n_epochs, not cfgs.eval_only)
if __name__ == "__main__":
main()