-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathHPPO_FetchReach_v1.py
173 lines (132 loc) · 6.27 KB
/
HPPO_FetchReach_v1.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
from HPPO.continuous import Agent
import gym
import numpy as np
import argparse
class main():
def __init__(self,args,env_name):
env = gym.make(env_name) # The wrapper encapsulates the gym env
num_achive = env.observation_space['achieved_goal'].shape[0]
num_desire = env.observation_space['desired_goal'].shape[0]
num_obs = env.observation_space['observation'].shape[0]
num_actions = env.action_space.shape[0]
num_states = num_obs + num_desire + num_achive
# args
args.num_actions = num_actions
args.num_states = num_states
args.num_goal = num_desire
# env
env = EnvPanda(env_name)
args.x_max = env.x_max
args.x_min = env.x_min
args.y_max = env.y_max
args.y_min = env.y_min
# print args
print("---------------")
for arg in vars(args):
print(arg,"=",getattr(args, arg))
print("---------------")
# create agent
hidden_layer_num_list = [256,256,256]
agent = Agent(args,hidden_layer_num_list)
#agent.load_actor_model("PPO_Actor_"+env_name+"_pre.pt")
print(agent.actor)
print(agent.critic)
print("---------------")
# training for 10 times
for i in range(10):
agent = Agent(args,hidden_layer_num_list)
agent.train(args,env,env_name)
#agent.train(args,env,env_name)
path_actor = "model/HPPO_Actor_"+env_name+".pt"
path_critic = "model/HPPO_Critic_"+env_name+".pt"
print("Saving actor model at:",path_actor)
print("Saving critic model at:",path_critic)
agent.save_actor_model(path_actor)
agent.save_critic_model(path_critic)
# save the normalization paramter to yaml
agent.state_norm.save_yaml(env_name+"_state.yaml")
agent.goal_norm.save_yaml(env_name+"_goal.yaml")
# evaluate
env_evaluate = EnvPanda(env_name,render_mode='human')
for i in range(10000):
evaluate_reward = agent.evaluate_policy(args, env_evaluate,render=True)
print("Evaluate reward:",evaluate_reward)
# overwrite env
class EnvPanda(gym.Env):
def __init__(self,env_name,render_mode=None):
if render_mode == None:
self.env = gym.make(env_name) # The wrapper encapsulates the gym env
self.render_mode = render_mode
else:
self.env = gym.make(env_name) # The wrapper encapsulates the gym env
self.render_mode = render_mode
self.threshold = 0.05
x = self.env.initial_gripper_xpos[0]
y = self.env.initial_gripper_xpos[1]
self.target_range = self.env.target_range
self.x_max = x + self.target_range
self.x_min = x - self.target_range
self.y_max = y + self.target_range
self.y_min = y - self.target_range
def distance(self,p1,p2):
d = (p1-p2)**2
d = np.sum(d)
d = np.sqrt(d)
return d
def compute_reward(self,ach,des):
costs = self.distance(ach,des)
reward = 50.0 if costs < self.threshold else 0
return reward
def step(self, action):
if self.render_mode == 'human':
self.env.render()
state, reward,truncted, info = self.env.step(action) # calls the gym env methods
ach = state['achieved_goal']
des = state['desired_goal']
obs = state['observation']
reward = self.compute_reward(ach,des)
bdw = True if reward > 0 else False
if reward == 0:
if self.distance(ach,self.pervious_ach) > 0.001:
reward = 1
self.pervious_ach = ach
state = np.hstack([des,ach,obs])
return state, reward, bdw , truncted, info
def reset(self):
while True:
obs = self.env.reset() # same for reset
#ach = self.object
ach = obs['achieved_goal']
des = obs['desired_goal']
obs = obs['observation']
if self.compute_reward(ach,des) > 0:
continue
else:
break
self.init_ach = ach
self.count = 1
self.pervious_ach = ach
state = np.hstack([des,ach,obs])
return [state]#[np.hstack([obs,des,ach])]
if __name__ == '__main__':
parser = argparse.ArgumentParser("Hyperparameters Setting for HPPO-continuous")
parser.add_argument("--lr", type=float, default=3e-4, help="Learning rate of actor")
parser.add_argument("--gamma", type=float, default=0.98, help="Discount factor")
parser.add_argument("--lamda", type=float, default=0.95, help="GAE parameter")
parser.add_argument("--epochs", type=int, default=5, help="HPPO training iteration parameter")
parser.add_argument("--epsilon", type=float, default=0.2, help="HPPO clip parameter")
parser.add_argument("--entropy_coef", type=float, default=0.00, help="Trick 5: policy entropy")
parser.add_argument("--max_train_steps", type=int, default=int(2e6), help=" Maximum number of training steps")
parser.add_argument("--max_rollout_step", type=int, default=3200, help=" Maximum number of rollout steps")
parser.add_argument("--use_hindsight_goal", type=bool, default=True, help="Flag for using hindsight goal")
parser.add_argument("--evaluate_freq", type=int, default=1, help="Evaluate the policy every 'evaluate_freq' steps")
parser.add_argument("--save_model_freq_training_epoch", type=int, default=10, help="Save model frequance")
parser.add_argument("--mini_batch_size_ratio", type=int, default=512, help="mini_batch_size_ratio")
parser.add_argument("--use_state_norm", type=bool, default=True, help="Flag for using state normalization")
parser.add_argument("--use_goal_norm", type=bool, default=True, help="Flag for using state normalization")
parser.add_argument("--use_HGF", type=bool, default=False, help="Flag for using hindsight goal filter")
parser.add_argument("--actor_std_min", type=float, default=1.1, help="Flag for using hindsight goal filter")
parser.add_argument("--env_name", type=str, default="FetchReach-v1", help=" Maximum number of rollout steps")
args = parser.parse_args()
env_name = args.env_name
main(args,env_name=env_name)