/
multiagentpicktest.py
148 lines (126 loc) · 3.84 KB
/
multiagentpicktest.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
#######version differer
import csv
import datetime
import numpy as np
import gym
import random
import MultiAgentEnv
import mujoco_py
import math
import tensorflow as tf
import tensorflow.contrib as tc
from actorpick import actor
from criticpick import critic
from collections import deque
env = gym.make('MultiAgentEnv-v0')
action_size = 4
state_size = 16
action_bound = env.action_space.high[:4]
print(action_bound)
batch_size = 128
import random
import matplotlib.pyplot as plt
###################seeding###################
seeding = 1234
np.random.seed(seeding)
tf.set_random_seed(seeding)
env.seed(seeding)
######################################
def cut_action_batch(batch):
batch2 = np.empty([batch_size,action_size])
for i in range(batch_size):
batch2[i] = batch[i][:4]
return batch2
#############This noise code is copied from openai baseline #########OrnsteinUhlenbeckActionNoise############# Openai Code#########
class OrnsteinUhlenbeckActionNoise:
def __init__(self, mu, sigma=0.3, theta=.15, dt=1e-2, x0=None):
self.theta = theta
self.mu = mu
self.sigma = sigma
self.dt = dt
self.x0 = x0
self.reset()
def __call__(self):
x = self.x_prev + self.theta * (self.mu - self.x_prev) * self.dt + \
self.sigma * np.sqrt(self.dt) * np.random.normal(size=self.mu.shape)
self.x_prev = x
return x
def reset(self):
self.x_prev = self.x0 if self.x0 is not None else np.zeros_like(self.mu)
def __repr__(self):
return 'OrnsteinUhlenbeckActionNoise(mu={}, sigma={})'.format(self.mu, self.sigma)
#########################################################################################################
def store_sample(s,a,r,d,info ,s2):
x = s['observation'][3:]
x2 = s2['observation'][3:]
ob_1 = np.reshape(x,(1,16))
ob_2 = np.reshape(x2,(1,16))
replay_memory.append((ob_1,a,r,d,ob_2))
def get_pick_reward(s):
if (s['achieved_goal'][2]) > 0.030:
r = 0
else:
r = -1
return r
def stg(s):
x = s['observation'][3:]
ob_1 = np.reshape(x,(1,16))
return ob_1
def compute_dist(achieved_goal, goal):
temp_desired = goal
temp_achieved = achieved_goal
eu_vector = []
for i in range(len(temp_desired)):
x = temp_desired[i]-temp_achieved[i]
x = x*x
eu_vector.append(x)
sqr_sum = sum(eu_vector)
sqrt = math.sqrt(sqr_sum)
return sqrt
sess = tf.Session()
ac = actor(state_size, action_size, action_bound, sess,ini=True)
cr = critic(state_size, action_size, action_bound, sess,ini=True)
s = env.reset()
noice = OrnsteinUhlenbeckActionNoise(mu=np.zeros(action_size*2))
scores_path = 'model/single_ddpg/scores/pickscore-{date:%Y-%m-%d %H:%M:%S}.csv'.format( date=datetime.datetime.now() )
demo_actions = np.load('demoactions.npy')
demo_actions_counter = 0
demo_pick_actions = np.load('demopick.npy')
demo_pick_actions_counter = 0
demo_pick_flag=False
sess.run(tf.global_variables_initializer())
save_path = 'model/multi_ddpg/'
saver = tf.train.Saver()
replay_memory = deque(maxlen = 100000)
max_ep = 50000
max_ep_len = 250
ac.lam=0.1
ac.update_ac=0
demo_ep_threshold=-1
pick_ep_threshold=-1
gamma = 0.99
R_graph = deque(maxlen = 10)
R_graph_= []
saver.restore(sess, "model/multi_ddpg_pick/pick_model.ckpt")
for ii in range(max_ep):
env = env.unwrapped
env.set_random_goal(True)
s = env.reset()
R,r = 0,0
for kk in range(max_ep_len):
ss = stg(s)
if ac.updateac>-1:
a = ac.get_actions_(ss)
b = a #+ noice()
b[3] = a[3]
b[7] = a[7]
a = b
else:
a = ac.get_action(ss)
a=a[0]
env.render()
s2,r,d,info=env.step(a)
ac.updateac=0
R_graph.append(R)
R_graph_.append(R)
ac.update_ac=0