-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathneural_net_for_hadron_1version2.py
191 lines (155 loc) · 6.94 KB
/
neural_net_for_hadron_1version2.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
# -*- coding: utf-8 -*-
import random
import gym
import numpy as np
from collections import deque
from keras.models import Sequential
from keras.layers import Dense
from keras.optimizers import Adam
import pybullet as p
import math
import pybullet_data
import time as t
EPISODES = 1000000
class DQNAgent:
def __init__(self, state_size, action_size):
self.state_size = state_size
self.action_size = action_size
self.memory = deque(maxlen=2000)
self.gamma = 0.95 # discount rate
self.epsilon = 1.0 # exploration rate
self.epsilon_min = 0.01
self.epsilon_decay = 0.995
self.learning_rate = 0.001
self.model = self._build_model()
def _build_model(self):
# Neural Net for Deep-Q learning Model
model = Sequential()
model.add(Dense(50, input_dim=self.state_size, activation='relu'))
model.add(Dense(50, activation='relu'))
model.add(Dense(50, activation='relu'))
model.add(Dense(50, activation='relu'))
model.add(Dense(self.action_size, activation='linear'))
model.compile(loss='mse',
optimizer=Adam(lr=self.learning_rate))
return model
def remember(self, state, action, reward, next_state, done):
self.memory.append((state, action, reward, next_state, done))
def act(self, state):
if np.random.rand() <= self.epsilon:
return random.randrange(self.action_size)
act_values = self.model.predict(state)
return np.argmax(act_values[0]) # returns action
def replay(self, batch_size):
minibatch = random.sample(self.memory, batch_size)
for state, action, reward, next_state, done in minibatch:
target = reward
if not done:
target = (reward + self.gamma *
np.amax(self.model.predict(next_state)[0]))
target_f = self.model.predict(state)
target_f[0][action] = target
self.model.fit(state, target_f, epochs=1, verbose=0)
if self.epsilon > self.epsilon_min:
self.epsilon *= self.epsilon_decay
def load(self, name):
self.model.load_weights(name)
def save(self, name):
self.model.save_weights(name)
if __name__ == "__main__":
#env = gym.make('CartPole-v1')
p.connect(p.GUI)
p.setAdditionalSearchPath(pybullet_data.getDataPath())
p.loadURDF("plane.urdf",0,0,0)
p.setGravity(0,0,0)
hadron = p.loadURDF("/home/arashrobo/Desktop/IRIS_2017/hadron_urdf/urdf/hadron_urdf.urdf")
p.setRealTimeSimulation(1)
#state_size = env.observation_space.shape[0]
state_size = 26
action_size = 32
agent = DQNAgent(state_size, action_size)
# agent.load("./save/cartpole-dqn.h5")
done = False
batch_size = 32
last = -1
for e in range(EPISODES):
#state = env.reset()
#reset
p.resetSimulation()
#p.loadURDF("plane.urdf")
p.loadSDF("stadium.sdf")
p.setGravity(0,0,-10)
hadron = p.loadURDF("/home/arashrobo/Desktop/IRIS_2017/hadron_urdf/urdf/hadron_urdf.urdf")
state = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
state[0] = p.getBasePositionAndOrientation(hadron)[0][0]
state[1] = p.getBasePositionAndOrientation(hadron)[0][1]
state[2] = p.getBasePositionAndOrientation(hadron)[0][2]
state[3] = p.getEulerFromQuaternion(p.getBasePositionAndOrientation(hadron)[1])[0]
state[4] = p.getEulerFromQuaternion(p.getBasePositionAndOrientation(hadron)[1])[1]
state[5] = p.getEulerFromQuaternion(p.getBasePositionAndOrientation(hadron)[1])[2]
for n in range (0,15):
state[6+n] = p.getJointState(hadron,n)[0]
print "state"
print(state[n])
state[25] = last
#reset
state = np.reshape(state, [1, state_size])
for time in range(1000):
#env.render()
action = agent.act(state)
last = action
if(action<16):
print(action)
p.setJointMotorControl2(hadron,action,p.VELOCITY_CONTROL,targetVelocity= 5.233333)
t.sleep(0.03333)
p.setJointMotorControl2(hadron,action,p.VELOCITY_CONTROL,targetVelocity= 0)
else:
print(action)
p.setJointMotorControl2(hadron,(action-16),p.VELOCITY_CONTROL,targetVelocity= -5.233333)
t.sleep(0.0333)
p.setJointMotorControl2(hadron,(action-16),p.VELOCITY_CONTROL,targetVelocity= 0)
p.stepSimulation()
next_state = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
next_state[0] = p.getBasePositionAndOrientation(hadron)[0][0]
next_state[1] = p.getBasePositionAndOrientation(hadron)[0][1]
next_state[2] = p.getBasePositionAndOrientation(hadron)[0][2]
next_state[3] = p.getEulerFromQuaternion(p.getBasePositionAndOrientation(hadron)[1])[0]
next_state[4] = p.getEulerFromQuaternion(p.getBasePositionAndOrientation(hadron)[1])[1]
next_state[5] = p.getEulerFromQuaternion(p.getBasePositionAndOrientation(hadron)[1])[2]
for n in range (0,15):
next_state[6+n] = p.getJointState(hadron,n)[0]
next_state[25] = last
x0 = p.getBasePositionAndOrientation(hadron)[0][0]
y0 = p.getBasePositionAndOrientation(hadron)[0][1]
x1 = 0
y1 = 0
reward = math.sqrt(((x1-x0)*(x1-x0))+((y1-y0)*(y1-y0)))
print ("reward=")
print (reward)
c = p.getEulerFromQuaternion(p.getBasePositionAndOrientation(hadron)[1])[0]
b = p.getEulerFromQuaternion(p.getBasePositionAndOrientation(hadron)[1])[1]
#print(b)
if ((c>0.785 or c<(-0.785) or b>0.785 or b<-0.785) and time>500):
print("hi")
reward = -100
else:
if(c>0.785 or c<(-0.785)):
reward = reward - math.sqrt(c*c)
else:
reward = reward + math.sqrt(c*c)
if(b>0.785 or b<(-0.785)):
reward = reward - math.sqrt(b*b)
else:
reward = reward + math.sqrt(c*c)
#
#reward = reward if reward>-3 else -10000
next_state = np.reshape(next_state, [1, state_size])
agent.remember(state, action, reward, next_state, done)
state = next_state
if reward<-1 and time> 500:
print("episode: {}/{}, score: {}, e: {:.2}"
.format(e, EPISODES, time, agent.epsilon))
break
agent.save("sa.h5")# version 2
if len(agent.memory) > batch_size:
agent.replay(batch_size)