-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
295 lines (251 loc) · 12.8 KB
/
main.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
283
284
285
286
287
288
289
290
291
292
from cgitb import reset
import datetime
import os
from re import A
import time
import argparse
import pickle
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
from environment import Environment
from task import Task
from dataset import Dataset
import transporter
import gnn_agent
import transporter_graph_agent
import conv_mlp
import cv2 as cv
import torch
import utils
MAX_ORDER = 3
os.environ["TF_CPP_MIN_LOG_LEVEL"]='3'
object_param ={"rope":{"action_dim":24,"env_type":0},
"ring":{"action_dim":32,"env_type":1},
"cloth":{"action_dim":16,"env_type":2}}
if __name__ == '__main__':
# Parse command line arguments.
parser = argparse.ArgumentParser()
#simulation params
parser.add_argument('-disp',default=False,type=bool)
parser.add_argument('-remote',default=1,type=int)
parser.add_argument('-max_action',default=20,type=int)
parser.add_argument('-task_type',default="rope_line",type=str)
#rope:rope_line, rope_l, rope_v, rope_n
#ring:ring_circle, ring_square, ring_move
#cloth:cloth_fold, cloth_flatten,multi_tasktrain
parser.add_argument('--agent', default='gnn_lin',type=str)
#local_gnn, global_gnn, transporter, transporter-goal,conv_mlp, transporter-graph
#gnn_mean, gnn_max, gnn_add
#transporter params
parser.add_argument('--num_rots', default=1, type=int)
parser.add_argument('--subsamp_g', action='store_true')
parser.add_argument('--crop_bef_q', default=0, type=int, help='CoRL paper used 1')
#dataset params
parser.add_argument('-data_dir', default='/data/dataset')
parser.add_argument('-eposide_num',default=1100,type=int) #10 100 1000
parser.add_argument('-train_sample_num',default=1000,type=int) #must less than eposide_num 1000
parser.add_argument('-test_sample_num',default=20,type=int) #100
parser.add_argument('-train_run',default=0,type=int) #random index
parser.add_argument('-mode',default="train",type=str) # train test
# only useful for gnn agent
parser.add_argument('-learning_rate', default=0.0001,type=float) #rope_line,rope_l,rope_v:0.001,rope_n:0.0001
parser.add_argument('-batch_size',default=128,type=int)
args = parser.parse_args()
args.time_id = time.strftime("%m_%d_%H:%M")
object_type = args.task_type[:(args.task_type.find('_'))]
action_dim = object_param[object_type]['action_dim']
env_type = object_param[object_type]['env_type']
num_train_iters = 40000 #5 before 40000
if "gnn" in args.agent:
num_train_iters = num_train_iters*10
#dataset initialization
dataset_dir = os.path.join(args.data_dir,args.task_type)
if not os.path.exists(dataset_dir):
os.makedirs(dataset_dir)
dataset = Dataset(dataset_dir,args.agent,args.remote)
total_data_list = list(range(args.eposide_num))
choose_episodes = np.random.choice(total_data_list, 1100, False)
train_episodes = choose_episodes[:args.train_sample_num]
dataset.set(train_episodes)
dataset.random_sample()
# Evaluate on increasing orders of magnitude of demonstrations.
test_dir = os.path.join(args.data_dir,'test_results')
if not os.path.exists(test_dir):
os.makedirs(test_dir)
check_point_dir = os.path.join(args.data_dir,'checkpoints')
if not os.path.exists(check_point_dir):
os.makedirs(check_point_dir)
log_dir = os.path.join(args.data_dir,'logs')
if not os.path.exists(log_dir):
os.makedirs(log_dir)
# Set the beginning of the agent name.
train_run = args.train_run
name = f'{args.task_type}-{args.agent}-{args.train_sample_num}-{train_run}'
if 'gnn' not in args.agent:
# GPU devices
physical_devices = tf.config.list_physical_devices('GPU')
for device in physical_devices:
tf.config.experimental.set_memory_growth(device, True)
# Set up tensorboard logger.
current_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
train_log_dir = os.path.join(log_dir, args.agent, args.task_type, current_time, 'train')
train_summary_writer = tf.summary.create_file_writer(train_log_dir)
# Initialize agent and limit random dataset sampling to fixed set.
tf.random.set_seed(train_run)
if args.agent == 'transporter':
agent = transporter.OriginalTransporterAgent(name,
args.task_type,
num_rotations=args.num_rots,
crop_bef_q=(args.crop_bef_q == 1))
elif 'transporter-goal' in args.agent:
agent = transporter.GoalTransporterAgent(name,
args.task_type,
num_rotations=args.num_rots)
elif 'gnn' in args.agent:
agent = gnn_agent.GNN_Agent(name,
args.task_type,
action_dim,
args.learning_rate,
args.batch_size,
args.agent)
elif args.agent == 'transporter-graph':
agent = transporter_graph_agent.Transporter_graph_Agent(
name,
args.task_type,
action_dim)
agent.models_dir = os.path.join(check_point_dir, name)
if not os.path.exists(agent.models_dir):
os.makedirs(agent.models_dir)
# Limit random data sampling to fixed set.
np.random.seed(train_run)
total_data_list = list(range(args.eposide_num))
choose_episodes = np.random.choice(total_data_list, 1100, False)
train_episodes = choose_episodes[:args.train_sample_num]
test_episodes = choose_episodes[-args.test_sample_num:]
if args.mode == "train":
dataset.set(train_episodes)
train_interval = int((num_train_iters)/10)
while agent.total_iter < num_train_iters:
# Train agent.
if "gnn" in args.agent:
agent.train(dataset,num_iter=train_interval)
else:
agent.train(dataset, num_iter=train_interval, writer=train_summary_writer)
elif args.mode == "test":
dataset.set(test_episodes)
test_num_iter_list = []
for fname in sorted(os.listdir(agent.models_dir)):
if 'attention' in fname:
train_num_iter = int(fname[15:-3])
test_num_iter_list.append(train_num_iter)
elif 'gnn' in fname:
train_num_iter = int(fname[fname.find('-') + 1:-3])
test_num_iter_list.append(train_num_iter)
elif 'pick' in fname:
train_num_iter = int(fname[10:-3])
test_num_iter_list.append(train_num_iter)
test_num_iter_list.sort()
test_num_iter = test_num_iter_list[len(test_num_iter_list)-1]
print("test iter:",test_num_iter)
load_sucess = False
if 'transporter' in args.agent:
tf.keras.backend.set_learning_phase(0)
agent.load(test_num_iter)
load_sucess = True
elif 'gnn' in args.agent:
agent.agent.eval()
agent.load(test_num_iter)
load_sucess = True
#train_interval = 100
#agent.test(dataset, num_iter=train_interval)
if load_sucess:
simu_env = Environment(disp=args.disp,env_type=env_type)
simu_env.reset()
simu_env.start()
done_list = []
act_num_list = []
for iepisode in test_episodes:
is_episode_sample = np.int32(dataset.episode_id) == iepisode
episode_samples = np.argwhere(is_episode_sample).squeeze().reshape(-1)
print("test task:",iepisode)
def load(iepisode, field):
field_path = os.path.join(dataset.path, field)
ep_fname = f'{iepisode:06d}-{len(episode_samples)}.pkl'
return pickle.load(open(os.path.join(field_path, ep_fname), 'rb'))
goal = {}
goal_index = len(episode_samples)
goal['color'] = load(iepisode, 'multi_rgb')[goal_index]
goal['depth'] = load(iepisode, 'depth')[goal_index]
goal_image = load(iepisode, 'tp_rgb')[goal_index]
goal_state = load(iepisode, 'next')[goal_index-1][action_dim:]
if 'cloth' in args.task_type:
goal_angle = utils.get_transform_angle(goal_state,args.task_type)
simu_env.remove_deform()
simu_env.reset(give_angle=goal_angle)
task_kwargs = {'task_type':args.task_type,
"normalize_length":simu_env.normalize_length,
"target_state":goal_state,
"transform_angle":simu_env.index_angle}
new_eposide_task = Task(**task_kwargs)
if "fold" not in args.task_type:
simu_env.random_initialize()
task_act_num= 0
done = False
for j in range(args.max_action):
obs_state =simu_env.get_current_state() #get observation
new_eposide_task.get_state_space(obs_state)
_,done = new_eposide_task.get_reward(obs_state)
if done:
break
else:
front_image,front_depth = simu_env.render(simu_env.camera_config[0])
left_image,left_depth = simu_env.render(simu_env.camera_config[1])
right_image,right_depth = simu_env.render(simu_env.camera_config[2])
tp_image,tp_depth = simu_env.render(simu_env.camera_config_up)
current_state,current_matrix = new_eposide_task.get_state_space(obs_state)
if "graph" in args.agent:
obs = {}
obs['curr'] = tp_image
obs['goal'] = goal_image
obs['kp_pos'] = current_state
elif "transporter" in args.agent:
obs = {}
obs['color'] = np.uint8([front_image,left_image,right_image])
obs['depth'] = np.float32([front_depth,left_depth,right_depth])
elif "gnn" in args.agent:
obs = current_state
if "goal" in args.agent:
act= agent.act(obs,goal)
else:
act= agent.act(obs)
#print("inference time:",extra["inference_time"])
if (args.agent == "transporter") or (args.agent == "transporter-goal"):
pick_position = [act['params']['pose0'][0],[0,0,0,1]]
place_position = [act['params']['pose1'][0],[0,0,0,1]]
simu_env.pick_place(pick_position,place_position)
else:
simu_env.step(act['params']['pose0'],act['params']['pose1'])
task_act_num += 1
print("task finish:",done)
tp_image_f,_ = simu_env.render(simu_env.camera_config_up)
for i in range(action_dim):
for p in range(5):
for q in range(5):
tp_image_f[new_eposide_task.target_state[i][1]+p][new_eposide_task.target_state[i][0]+q] = [255,0,0]
cv.imwrite("resu_"+str(len(done_list))+".jpg",tp_image_f)
done_list.append(done)
act_num_list.append(task_act_num)
simu_env.remove_deform()
simu_env.reset()
print("sucessful_rate:",np.sum(done_list)/args.test_sample_num)
test_dic = {}
test_dic["successful"] = done_list
test_dic["act_num"] = act_num_list
test_dic["iter"] = test_num_iter
pickle_fname = os.path.join(test_dir, f'{name+str(test_num_iter)}.pkl')
pickle.dump(test_dic, open(pickle_fname, 'wb'))
print("dump:",name)
simu_env.pause()
simu_env.stop()
del simu_env