-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathPI.py
408 lines (331 loc) · 21.3 KB
/
PI.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
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
# -*- coding: UTF-8 -*-
import vrep
import numpy
import sys, time, threading
from math import *
roll_outs = 30 # 每一个时间步随机产生的轨迹数
training_times = 60 # 训练次数
N = 8
class PI2:
def __init__(self):
self.start = 0, 0 # 起始点位置
self.goal = 1.5, 1.5 # 目标点位置，米
self.L = sqrt(pow(self.goal[0] - self.start[0], 2) + pow(self.goal[1] - self.start[1], 2)) # 距离
self.motion_area_point1 = -0.5, -0.5
self.motion_area_point2 = 2.5, 2.5 # 运动区域
self.max_cost = sqrt(pow(self.motion_area_point1[0] - self.motion_area_point2[0], 2) + pow(
self.motion_area_point1[1] - self.motion_area_point2[1], 2))
self.constant1 = 30 # 常数1
self.T = 12 # 运行时间/s
self.T1 = 6.0 # 中间点时间
self.dt = 0.05 # 控制周期/s
self.current_time = 0 # 当前时刻
self.current_roll_out = 0 # 当前轨迹号
self.current_train = 0 # 当前训练次数
self.end_time_position = [] # 终止时刻位置，三维数据,x,y,z
self.init_obstacle1_position = 0.25, 0.7, 0.05
self.init_obstacle2_position = 0.95, 0.4, 0.05
self.obstacle1_position = []
self.obstacle2_position = []
# 弧度制
self.joint_angle = numpy.zeros((N, 1), dtype=numpy.float64) # 关节角度
# snake_A = 0.592
# snake_omega = 6.361
# snake_beta = 1.046
# snake_gamma = -0.064
# 蛇形曲线参数
self.A1 = 0.592
self.omega1 = 2 * pi # [0.5pi,4pi]
self.beta1 = 1.046
self.gamma1 = 0 # [-0.12,0.12]
self.A2 = 0.592
self.omega2 = 2 * pi # [0.5pi,4pi]
self.beta2 = 1.046
self.gamma2 = 0 # [-0.12,0.12]
self.A01 = 0.592
self.omega01 = 2 * pi # [0.5pi,4pi]
self.beta01 = 1.046
self.gamma01 = 0.06 # [-0.12,0.12]
self.A02 = 0.592
self.omega02 = 2 * pi # [0.5pi,4pi]
self.beta02 = 1.046
self.gamma02 = 0.06 # [-0.12,0.12]
# 正态分布标准差
self.omega_gamma1 = 0.5
self.gamma_gamma1 = 0.02
self.omega_gamma2 = 0.5
self.gamma_gamma2 = 0.02
# 每条轨迹的训练参数变化量
self.omega_delta1 = numpy.zeros(roll_outs, dtype=numpy.float64)
self.gamma_delta1 = numpy.zeros(roll_outs, dtype=numpy.float64)
self.omega_delta2 = numpy.zeros(roll_outs, dtype=numpy.float64)
self.gamma_delta2 = numpy.zeros(roll_outs, dtype=numpy.float64)
# 每条轨迹的训练参数
self.omega_roll1 = self.omega2 * numpy.ones(roll_outs, dtype=numpy.float64)
self.gamma_roll1 = self.gamma2 * numpy.ones(roll_outs, dtype=numpy.float64)
self.omega_roll2 = self.omega2 * numpy.ones(roll_outs, dtype=numpy.float64)
self.gamma_roll2 = self.gamma2 * numpy.ones(roll_outs, dtype=numpy.float64)
# 所有的训练参数
self.omega_all1 = numpy.zeros((training_times, roll_outs), dtype=numpy.float64)
self.gamma_all1 = numpy.zeros((training_times, roll_outs), dtype=numpy.float64)
self.omega_all2 = numpy.zeros((training_times, roll_outs), dtype=numpy.float64)
self.gamma_all2 = numpy.zeros((training_times, roll_outs), dtype=numpy.float64)
# 每次训练后的损失函数
self.cost_train = numpy.zeros(training_times, dtype=numpy.float64)
self.cost_all = numpy.zeros((training_times, roll_outs), dtype=numpy.float64)
self.cost = numpy.zeros(roll_outs, dtype=numpy.float64)
# 每次训练后通过PI2计算的参数
self.omega_train1 = self.omega2 * numpy.ones(training_times, dtype=numpy.float64)
self.gamma_train1 = self.gamma2 * numpy.ones(training_times, dtype=numpy.float64)
self.omega_train2 = self.omega2 * numpy.ones(training_times, dtype=numpy.float64)
self.gamma_train2 = self.gamma2 * numpy.ones(training_times, dtype=numpy.float64)
#每次训练通过PI2计算的参数，继续产生五条轨迹，取最小损失函数，作为每次迭代后的损失函数
self.PI2_cost = numpy.zeros(5, dtype=numpy.float64)
self.PI2_counts = 0
def sim_start(self):
print('开始仿真')
vrep.simxFinish(-1) # 关闭所有可能的连接
self.clientID = vrep.simxStart('127.0.0.1', 19997, True, True, 5000, 5) # 创建和vrep之间的通信线程
vrep.simxStartSimulation(self.clientID, vrep.simx_opmode_blocking) # 开始仿真
res, self.HandleObstacle1 = vrep.simxGetObjectHandle(self.clientID, "obstacle1", vrep.simx_opmode_oneshot_wait)
res, self.HandleObstacle2 = vrep.simxGetObjectHandle(self.clientID, "obstacle2", vrep.simx_opmode_oneshot_wait)
res, self.HandleJoint1 = vrep.simxGetObjectHandle(self.clientID, "yaw_joint0", vrep.simx_opmode_oneshot_wait)
res, self.HandleJoint2 = vrep.simxGetObjectHandle(self.clientID, "yaw_joint1", vrep.simx_opmode_oneshot_wait)
res, self.HandleJoint3 = vrep.simxGetObjectHandle(self.clientID, "yaw_joint2", vrep.simx_opmode_oneshot_wait)
res, self.HandleJoint4 = vrep.simxGetObjectHandle(self.clientID, "yaw_joint3", vrep.simx_opmode_oneshot_wait)
res, self.HandleJoint5 = vrep.simxGetObjectHandle(self.clientID, "yaw_joint4", vrep.simx_opmode_oneshot_wait)
res, self.HandleJoint6 = vrep.simxGetObjectHandle(self.clientID, "yaw_joint5", vrep.simx_opmode_oneshot_wait)
res, self.HandleJoint7 = vrep.simxGetObjectHandle(self.clientID, "yaw_joint6", vrep.simx_opmode_oneshot_wait)
res, self.HandleJoint8 = vrep.simxGetObjectHandle(self.clientID, "yaw_joint7", vrep.simx_opmode_oneshot_wait)
res, self.HandleHead = vrep.simxGetObjectHandle(self.clientID, "snake", vrep.simx_opmode_oneshot_wait)
for training_time in range(training_times):
self.current_train = training_time # 当前训练次数
for roll_out in range(roll_outs):
self.current_roll_out = roll_out # 当前轨迹数
# 产生随机数，训练参数的变化量
self.omega_delta1[roll_out] = numpy.random.normal(0.0, self.omega_gamma1, 1)
self.omega_delta2[roll_out] = numpy.random.normal(0.0, self.omega_gamma2, 1)
self.gamma_delta1[roll_out] = numpy.random.normal(0.0, self.gamma_gamma1, 1)
self.gamma_delta2[roll_out] = numpy.random.normal(0.0, self.gamma_gamma2, 1)
# 更新训练参数
self.omega_roll1[roll_out] = self.omega01 + self.omega_delta1[roll_out]
self.omega_roll2[roll_out] = self.omega02 + self.omega_delta2[roll_out]
self.gamma_roll1[roll_out] = self.gamma01 + self.gamma_delta1[roll_out]
self.gamma_roll2[roll_out] = self.gamma02 + self.gamma_delta2[roll_out]
# 训练参数限幅
if self.omega_roll1[roll_out] > 4 * pi:
self.omega_roll1[roll_out] = 4 * pi
if self.omega_roll1[roll_out] < 0.5 * pi:
self.omega_roll1[roll_out] = 0.5 * pi
if self.omega_roll2[roll_out] > 4 * pi:
self.omega_roll2[roll_out] = 4 * pi
if self.omega_roll2[roll_out] < 0.5 * pi:
self.omega_roll2[roll_out] = 0.5 * pi
if self.gamma_roll1[roll_out] > 0.12:
self.gamma_roll1[roll_out] = 0.12
if self.gamma_roll1[roll_out] < -0.12:
self.gamma_roll1[roll_out] = -0.12
if self.gamma_roll1[roll_out] > 0:
self.gamma_roll1[roll_out] = - self.gamma_roll1[roll_out]
if self.gamma_roll2[roll_out] > 0.12:
self.gamma_roll2[roll_out] = 0.12
if self.gamma_roll2[roll_out] < -0.12:
self.gamma_roll2[roll_out] = -0.12
if self.gamma_roll2[roll_out] > 0:
self.gamma_roll2[roll_out] = - self.gamma_roll2[roll_out]
# 记录每一条轨迹的随机参数
self.omega_all1[self.current_train, self.current_roll_out] = self.omega_roll1[roll_out]
self.omega_all2[self.current_train, self.current_roll_out] = self.omega_roll2[roll_out]
self.gamma_all1[self.current_train, self.current_roll_out] = self.gamma_roll1[roll_out]
self.gamma_all2[self.current_train, self.current_roll_out] = self.gamma_roll2[roll_out]
print(self.omega_roll1[roll_out], self.omega_roll2[roll_out], self.gamma_roll1[roll_out],
self.gamma_roll2[roll_out])
vrep.simxFinish(-1) # 关闭所有可能的连接
self.clientID = vrep.simxStart('127.0.0.1', 19997, True, True, 5000, 5) # 创建和vrep之间的通信线程
vrep.simxStartSimulation(self.clientID, vrep.simx_opmode_blocking) # 开始仿真
self.time_step()
time.sleep(21)
exponential_value_cost = numpy.zeros(roll_outs, dtype=numpy.float64) # 代价指数
probability_weighting = numpy.zeros(roll_outs, dtype=numpy.float64) # 每条轨迹的概率
for i in range(roll_outs):
exponential_value_cost[i] = exp(
-self.constant1 * (self.cost[i] - self.cost.min()) / (self.cost.max() - self.cost.min()))
for i in range(roll_outs):
probability_weighting[i] = exponential_value_cost[i] / numpy.sum(exponential_value_cost)
temp_delta_omega1 = 0
temp_delta_omega2 = 0
temp_delta_gamma1 = 0
temp_delta_gamma2 = 0
for i in range(roll_outs):
temp_delta_omega1 = temp_delta_omega1 + probability_weighting[i] * self.omega_delta1[i]
temp_delta_omega2 = temp_delta_omega2 + probability_weighting[i] * self.omega_delta2[i]
temp_delta_gamma1 = temp_delta_gamma1 + probability_weighting[i] * self.gamma_delta1[i]
temp_delta_gamma2 = temp_delta_gamma2 + probability_weighting[i] * self.gamma_delta2[i]
# 更新参数
self.omega01 = self.omega01 + temp_delta_omega1
self.omega_train1[self.current_train] = self.omega01
self.omega02 = self.omega02 + temp_delta_omega2
self.omega_train2[self.current_train] = self.omega02
self.gamma01 = self.gamma01 + temp_delta_gamma1
self.gamma01 = -abs(self.gamma01) # gamma1恒为负值
self.gamma_train1[self.current_train] = self.gamma01
self.gamma02 = self.gamma02 + temp_delta_gamma2
self.gamma02 = -abs(self.gamma02)
self.gamma_train2[self.current_train] = self.gamma02
# 计算每次迭代后的损失函数
for kk in range(5):
self.PI2_counts = kk
vrep.simxFinish(-1) # 关闭所有可能的连接
self.clientID = vrep.simxStart('127.0.0.1', 19997, True, True, 5000, 5) # 创建和vrep之间的通信线程
vrep.simxStartSimulation(self.clientID, vrep.simx_opmode_blocking) # 开始仿真
self.time_step_train()
time.sleep(21)
self.cost_train[self.current_train] = self.PI2_cost.min()
if self.cost_train[self.current_train] >= 0.5:
self.omega_gamma1 = 0.5
self.omega_gamma2 = 0.5
self.gamma_gamma1 = 0.02
self.gamma_gamma2 = 0.02
if self.cost_train[self.current_train] < 0.5 and self.cost_train[self.current_train] >= 0.1:
self.omega_gamma2 = self.cost_train[self.current_train] * 0.5
self.omega_gamma2 = self.cost_train[self.current_train] * 0.5
self.gamma_gamma1 = self.cost_train[self.current_train] * 0.02
self.gamma_gamma2 = self.cost_train[self.current_train] * 0.02
if self.cost_train[self.current_train] < 0.1:
self.omega_gamma1 = self.cost_train[self.current_train] * 0.5 / 2
self.omega_gamma2 = self.cost_train[self.current_train] * 0.5 / 2
self.gamma_gamma1 = self.cost_train[self.current_train] * 0.02 / 2
self.gamma_gamma2 = self.cost_train[self.current_train] * 0.02 / 2
def get_positon_from_vrep(self): # 获取头部位置
res, self.end_time_position = vrep.simxGetObjectPosition(self.clientID, self.HandleHead, -1,
vrep.simx_opmode_oneshot_wait) # 从vrep仿真环境下获取头部实时位置
def satisfy_constraint(self):
res, self.obstacle1_position = vrep.simxGetObjectPosition(self.clientID, self.HandleObstacle1, -1,
vrep.simx_opmode_oneshot_wait) # 从vrep仿真环境下获取头部实时位置
res, self.obstacle2_position = vrep.simxGetObjectPosition(self.clientID, self.HandleObstacle2, -1,
vrep.simx_opmode_oneshot_wait) # 从vrep仿真环境下获取头部实时位置
temp1 = abs(self.obstacle1_position[0] - self.init_obstacle1_position[0])
temp2 = abs(self.obstacle1_position[1] - self.init_obstacle1_position[1])
temp3 = abs(self.obstacle1_position[2] - self.init_obstacle1_position[2])
temp4 = abs(self.obstacle2_position[0] - self.init_obstacle2_position[0])
temp5 = abs(self.obstacle2_position[1] - self.init_obstacle2_position[1])
temp6 = abs(self.obstacle2_position[2] - self.init_obstacle2_position[2])
if temp1 > 0.005 or temp2 > 0.005 or temp3 > 0.005 or temp4 > 0.005 or temp5 > 0.005 or temp6 > 0.005:
return False
temp1 = self.motion_area_point1[0] - self.end_time_position[0]
temp2 = self.motion_area_point1[1] - self.end_time_position[1]
temp3 = self.end_time_position[0] - self.motion_area_point2[0]
temp4 = self.end_time_position[1] - self.motion_area_point2[1]
if temp1 > 0 or temp2 > 0 or temp3 > 0 or temp4 > 0:
return False
return True
def time_step_train(self):
if self.current_time >= self.T: # 到达终止时刻
self.get_positon_from_vrep() # 获取终止时刻头部位置
self.PI2_cost[self.PI2_counts] = sqrt(pow(self.goal[0] - self.end_time_position[0], 2) + pow(self.goal[1] - self.end_time_position[1],2)) # 计算代价
# 越界或者碰到障碍物
flag_temp = self.satisfy_constraint()
if flag_temp == False:
self.PI2_cost[self.PI2_counts] = self.max_cost
print(self.PI2_cost[self.PI2_counts])
##在vrep里让机器人立马回到初始位姿
vrep.simxStopSimulation(self.clientID, vrep.simx_opmode_blocking)
vrep.simxFinish(self.clientID)
self.current_time = 0
return
# 计算每个时间步关节角度
for i in range(N):
if self.current_time < self.T1:
self.joint_angle[N - 1 - i] = -self.A1 * sin(
self.omega_train1[self.current_train] * self.current_time + (i - 1) * self.beta1) + \
self.gamma_train1[self.current_train]
if self.current_time >= self.T1:
self.joint_angle[N - 1 - i] = -self.A2 * sin(
self.omega_train2[self.current_train] * self.current_time + (i - 1) * self.beta2) + \
self.gamma_train2[self.current_train]
# 关节角度限幅
if self.joint_angle[i] > 60 / 180 * pi:
self.joint_angle[i] = 60 / 180 * pi
if self.joint_angle[i] < -60 / 180 * pi:
self.joint_angle[i] = -60 / 180 * pi
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint1, self.joint_angle[0],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint2, self.joint_angle[1],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint3, self.joint_angle[2],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint4, self.joint_angle[3],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint5, self.joint_angle[4],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint6, self.joint_angle[5],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint7, self.joint_angle[6],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint8, self.joint_angle[7],
vrep.simx_opmode_oneshot)
self.current_time += self.dt
timer = threading.Timer(self.dt, self.time_step_train)
timer.start()
def time_step(self):
if self.current_time >= self.T: # 到达终止时刻
self.get_positon_from_vrep() # 获取终止时刻头部位置
self.cost[self.current_roll_out] = sqrt(pow(self.goal[0] - self.end_time_position[0], 2) + pow(self.goal[1] - self.end_time_position[1],2)) # 计算代价
# 越界或者碰到障碍物
flag_temp = self.satisfy_constraint()
if flag_temp == False:
self.cost[self.current_roll_out] = self.max_cost
print(self.cost[self.current_roll_out])
##在vrep里让机器人立马回到初始位姿
vrep.simxStopSimulation(self.clientID, vrep.simx_opmode_blocking)
vrep.simxFinish(self.clientID)
self.current_time = 0
return
# 计算每个时间步关节角度
for i in range(N):
if self.current_time < self.T1:
self.joint_angle[N - 1 - i] = -self.A1 * sin(
self.omega_roll1[self.current_roll_out] * self.current_time + (i - 1) * self.beta1) + \
self.gamma_roll1[self.current_roll_out]
if self.current_time >= self.T1:
self.joint_angle[N - 1 - i] = -self.A2 * sin(
self.omega_roll2[self.current_roll_out] * self.current_time + (i - 1) * self.beta2) + \
self.gamma_roll2[self.current_roll_out]
# 关节角度限幅
if self.joint_angle[i] > 60 / 180 * pi:
self.joint_angle[i] = 60 / 180 * pi
if self.joint_angle[i] < -60 / 180 * pi:
self.joint_angle[i] = -60 / 180 * pi
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint1, self.joint_angle[0],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint2, self.joint_angle[1],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint3, self.joint_angle[2],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint4, self.joint_angle[3],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint5, self.joint_angle[4],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint6, self.joint_angle[5],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint7, self.joint_angle[6],
vrep.simx_opmode_oneshot)
vrep.simxSetJointTargetPosition(self.clientID, self.HandleJoint8, self.joint_angle[7],
vrep.simx_opmode_oneshot)
self.current_time += self.dt
timer = threading.Timer(self.dt, self.time_step)
timer.start()
if __name__ == '__main__':
mPI2 = PI2()
mPI2.sim_start()
# 保存数据
numpy.save('omega_all1.npy', mPI2.omega_all1) # n_trains * n_rolls
numpy.save('gamma_all1.npy', mPI2.gamma_all1) # n_trains * n_rolls
numpy.save('omega_all2.npy', mPI2.omega_all2) # n_trains * n_rolls
numpy.save('gamma_all2.npy', mPI2.gamma_all2) # n_trains * n_rolls
numpy.save('omega_train1.npy', mPI2.omega_train1) # n_trains
numpy.save('gamma_train1.npy', mPI2.gamma_train1) # n_trains
numpy.save('omega_train2.npy', mPI2.omega_train2) # n_trains
numpy.save('gamma_train2.npy', mPI2.gamma_train2) # n_trains
numpy.save('cost_all.npy', mPI2.cost_all) # n_trains * n_rolls
numpy.save('cost_train.npy', mPI2.cost_train) # n_trains