-
Notifications
You must be signed in to change notification settings - Fork 0
/
panda_sac.py
187 lines (161 loc) · 7.98 KB
/
panda_sac.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
from stable_baselines3 import HerReplayBuffer, SAC
from sb3_contrib import ARS, QRDQN, TQC, TRPO, RecurrentPPO
import gymnasium
import panda_gym
import numpy as np
from typing import Any, ClassVar, Dict, Iterable, List, Optional, Tuple, Type, TypeVar, Union
from stable_baselines3.common.noise import ActionNoise, VectorizedActionNoise
from gymnasium import spaces
from stable_baselines3.common.vec_env import (
DummyVecEnv,
unwrap_vec_normalize,
)
from stable_baselines3.common.evaluation import evaluate_policy
from model import SACEnvSwitchWrapper
from utils import get_push_env, get_pick_and_place_env, get_push_dense_env, get_push_joints_env, get_push_joints_dense_env, get_reach_env, get_slide_env, get_stack_env
import argparse
import numpy
def make_env(name,lateral_friction=1.0,spinning_friction=0.001,mass=1.0):
print(name)
if name == "PandaPush-v3":
print("This is PandaPush-v3 Env, Welcome!")
return get_push_env(lateral_friction, spinning_friction, mass)
elif name == "PandaPickAndPlace-v3":
print("This is PandaPickAndPlace-v3 Env, Welcome!")
return get_pick_and_place_env(lateral_friction, spinning_friction, mass)
elif name == "PandaPushDense-v3":
print("This is PandaPushDense-v3 Env, Welcome!")
return get_push_dense_env(lateral_friction, spinning_friction, mass)
elif name == "PandaPushJoints-v3":
print("This is PandaPushJoints-v3 Env, Welcome!")
return get_push_joints_env(lateral_friction, spinning_friction, mass)
elif name == "PandaPushJointsDense-v3":
print("This is PandaPushJointsDense-v3 Env, Welcome!")
return get_push_joints_dense_env(lateral_friction, spinning_friction, mass)
elif name == "PandaReach-v3":
print("This is PandaReach-v3 Env, Welcome!")
return get_reach_env(lateral_friction, spinning_friction, mass)
elif name == "PandaSlide-v3":
print("This is PandaSlide-v3 Env, Welcome!")
return get_slide_env(lateral_friction, spinning_friction, mass)
elif name == "PandaStack-v3":
print("This is PandaStack-v3 Env, Welcome!")
return get_stack_env(lateral_friction, spinning_friction, mass)
else:
print("Unkown Environment!! ")
def train(args):
env_id = args.domain_name
# log_dir = './panda_push_v3_tensorboard/'
log_dir = './' + args.domain_name + '_tensorboard/'
if args.test_mass == 1: # friction experiment
if args.test_lateral_friction == 1.0: # spinning_friction
env_int = numpy.random.uniform(low=args.random_float[0], high=args.random_float[1], size=4)
print("Random Spinning Friciton Values:", env_int)
# env for train
env1 = make_env(env_id, 1.0, env_int[0], 1.0)
env2 = make_env(env_id, 1.0, env_int[1], 1.0)
env3 = make_env(env_id, 1.0, env_int[2], 1.0)
env4 = make_env(env_id, 1.0, env_int[3], 1.0)
else:
env_int = numpy.random.uniform(low=1.0, high=4.0, size=4)
print("Random Lateral Friciton Values:", env_int)
# env for train
env1 = make_env(env_id, env_int[0], 0.001, 1.0)
env2 = make_env(env_id, env_int[1], 0.001, 1.0)
env3 = make_env(env_id, env_int[2], 0.001, 1.0)
env4 = make_env(env_id, env_int[3], 0.001, 1.0)
else:
env_int = numpy.random.randint(low=args.random_int[0], high=args.random_int[1], size=4, dtype='l')
print("Random Mass Values:", env_int)
# env for train
env1 = make_env(env_id, 1.0, 0.001, env_int[0])
env2 = make_env(env_id, 1.0, 0.001, env_int[1])
env3 = make_env(env_id, 1.0, 0.001, env_int[2])
env4 = make_env(env_id, 1.0, 0.001, env_int[3])
# env for test
env5 = make_env(env_id, args.test_lateral_friction, args.test_spinning_friction, args.test_mass)
train_env = DummyVecEnv([env1,env2,env3,env4])
test_env = DummyVecEnv([env5,env5,env5,env5])
# SAC train model
model = SACEnvSwitchWrapper(policy = "MultiInputPolicy",
env = train_env,
batch_size=2048,
gamma=0.95,
learning_rate=1e-4,
train_freq=64,
gradient_steps=64,
tau=0.05,
replay_buffer_class=HerReplayBuffer,
replay_buffer_kwargs=dict(
n_sampled_goal=4,
goal_selection_strategy="future",
),
policy_kwargs=dict(
net_arch=[512, 512, 512],
n_critics=2,
),
learning_starts = 1000,
verbose=1,
tensorboard_log=log_dir)
model.learn(total_timesteps=args.time_step,progress_bar=True)
train_env.close()
# save_model
if args.test_mass == 1: # friction experiment
if args.test_lateral_friction == 1.0: # spinning_friction
# save_model
model_name = "SAC-" + str(args.domain_name + "-model-spinning-friction" + str(args.test_spinning_friction))
model.save(model_name)
model.save_replay_buffer('SAC-' + str(args.domain_name) + '-buffer' + "-model-spinning-friction" + str(args.test_spinning_friction))
else:
# save_model
model_name = "SAC-" + str(args.domain_name + "-model-lateral-friction" + str(args.test_lateral_friction))
model.save(model_name)
model.save_replay_buffer('SAC-' + str(args.domain_name) + '-buffer' + "-model-lateral-friction" + str(args.test_lateral_friction))
else:
# save_model
model_name = "SAC-" + str(args.domain_name + "-model-mass" + str(args.test_mass))
model.save(model_name)
model.save_replay_buffer('SAC-' + str(args.domain_name) + '-buffer' + "-model-mass" + str(args.test_mass))
# train_env.save(f"{model_name}_vec_normalize.pkl")
# switch environment
model2 = SACEnvSwitchWrapper.load(model_name, env=test_env)
model2.eval_env = True
model2.learn(total_timesteps=args.time_step,progress_bar=True)
test_env.close()
def test(args):
# env for train
env1 = make_env()
env2 = make_env(env_id, 1.0, 0.001,2.0)
env3 = make_env(env_id, 1.0, 0.001,3.0)
env4 = make_env(env_id, 1.0, 0.001,4.0)
# env for test
env5 = make_env(env_id, 1.0, 0.001,10)
train_env = DummyVecEnv([env1,env2,env3,env4])
test_env = DummyVecEnv([env5,env5,env5,env5])
model = SACEnvSwitchWrapper.load('SAC-PandaPush-v3',env=train_env)
# model.env = train_env
# train_mean_reward, train_std_reward = evaluate_policy(model, train_env, 100)
test_mean_reward, test_std_reward = evaluate_policy(model, test_env, 100)
# print(f"Train Mean reward = {train_mean_reward:.2f} +/- {train_std_reward:.2f}")
print(f"Test Mean reward = {test_mean_reward:.2f} +/- {test_std_reward:.2f}")
train_env.close()
test_env.close()
def retrain(args):
# env for test
env5 = make_env(env_id, 1.0, 0.001,50)
test_env = DummyVecEnv([env5,env5,env5,env5])
model = SACEnvSwitchWrapper.load('SAC-PandaPush-v3',env=test_env)
model.eval_env = True
model.learn(total_timesteps=args.time_step,progress_bar=True)
test_env.close()
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--domain_name', default='PandaPush-v3')
parser.add_argument('--random_int', default=[1, 5], nargs='+', type=int)
parser.add_argument('--random_float', default=[0.001, 0.01], nargs='+', type=float)
parser.add_argument('--test_mass', default=1.0, type=int)
parser.add_argument('--time_step', default=800000, type=int)
parser.add_argument('--test_spinning_friction', default=0.001, type=float)
parser.add_argument('--test_lateral_friction', default=1.0, type=float)
args = parser.parse_args()
train(args)