-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtest_on_real_robot.py
217 lines (202 loc) · 13 KB
/
test_on_real_robot.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
import sys
sys.path.append("./utils")
try:
sys.path.remove('/opt/ros/kinetic/lib/python2.7/dist-packages')
except:
pass
import time
import os
import random
import matplotlib.pyplot as plt
import numpy as np
import cv2
import torch
from torch.autograd import Variable
from robot import Robot
from logger import Logger
from robot_stone import Robot
import heightmap
import math3d as m3d
import random
from math import *
import scipy
from torch import nn
from PIL import Image
import tier1_module
import tier2_module
import tier3_module
from datetime import datetime
from random import choice
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
htmap_w = 200
htmap_h = 200
workspace_limits_raw = np.array([[-0.015, 0.115], [0.63, 0.76], [0.02, 0.1]])
workspace_limits_list = [np.array([[-0.015, 0.05], [0.63, 0.695], [0.02, 0.1]]), np.array([[0.0175, 0.0825], [0.63, 0.695], [0.02, 0.1]]), np.array([[0.05, 0.115], [0.63, 0.695], [0.015, 0.1]]), np.array([[-0.015, 0.05], [0.6625, 0.7275], [0.02, 0.1]]), np.array([[0.0175, 0.0825], [0.6625, 0.7275], [0.02, 0.1]]), np.array([[0.05, 0.115], [0.6625, 0.7275], [0.02, 0.1]]), np.array([[-0.015, 0.05], [0.695, 0.76], [0.02, 0.1]]), np.array([[0.0175, 0.0825], [0.695, 0.76], [0.02, 0.1]]), np.array([[0.05, 0.115], [0.695, 0.76], [0.02, 0.1]])]
heightmap_resolution = 0.0013/4.0
net = tier1_module.Tier1(pcpt_block=tier1_module.BasicBlock, pcpt_layers=[1,5,1], scoop_block=tier1_module.BasicBlock, scoop_layers=[1,5,1], h=htmap_h, w=htmap_w).cuda() # define the network
net.load_state_dict(torch.load('netparam_Tier1.pkl')) # domino
net_submodule2 = tier2_module.Tier2(pcpt_block=tier2_module.BasicBlock, pcpt_layers=[1,1,1], scoop_block=tier2_module.BasicBlock, scoop_layers=[1,1,1], h=200, w=200).cuda()
net_submodule2.load_state_dict(torch.load('netparam_Tier2.pkl'))
net_submodule3 = tier3_module.Tier3(pcpt_block=tier3_module.BasicBlock, pcpt_layers=[1,5,1], scoop_block=tier3_module.BasicBlock, scoop_layers=[1,5,1], h=200, w=200).cuda()
net_submodule3.load_state_dict(torch.load('netparam_Tier3.pkl'))
def point_position_after_rotation(current_xy, rotation_pole, desired_angle):
desired_angle_rad = desired_angle*pi/180
current_displacement = list(np.array(current_xy)-np.array(rotation_pole))
rotation_matrix = [[cos(desired_angle_rad), -sin(desired_angle_rad)],[sin(desired_angle_rad), cos(desired_angle_rad)]]
current_displacement = np.expand_dims(current_displacement, axis=1)
temp = np.dot(rotation_matrix, current_displacement)
xy_after_rotate = [list(rotation_pole[0]+temp[0])[0], list(rotation_pole[1]+temp[1])[0]]
return xy_after_rotate
def from_angle_index_to_gripper_rot_z(angle_index):
if angle_index>=0 and angle_index<=8:
rot_z = (-pi/8)*angle_index
else:
rot_z = pi/8*(16-angle_index)
return rot_z
def from_pixel_to_world_position(pixel, workspace_limits, heightmap_resolution, depth_heightmap):
htmap_h = int(round((workspace_limits[1][1]-workspace_limits[1][0])/heightmap_resolution))
htmap_w = int(round((workspace_limits[0][1]-workspace_limits[0][0])/heightmap_resolution))
pix_x = pixel[0]
pix_y = pixel[1]
pos = [pix_x * heightmap_resolution + workspace_limits[0][0], pix_y * heightmap_resolution + workspace_limits[1][0], depth_heightmap[pix_y][pix_x] + workspace_limits[2][0]]
return pos
def main():
total_index = 0
robot = Robot("192.168.1.102")
while True:
while True:
whether_continue = input('Whether continue? (y or n) Shuffle!')
if whether_continue=='y':
break
max_score_pixel_list = []
color_img_array, depth_array = robot.getCameraData()
for workspace_limits in workspace_limits_list:
#color_heightmap, depth_heightmap = heightmap.get_heightmap(color_img_array, depth_array, cam_intrinsics, baseTcam, workspace_limits, heightmap_resolution)
color_heightmap, depth_heightmap = heightmap.get_heightmap(color_img_array, depth_array, robot.cam_intrinsics, robot.baseTcam, workspace_limits, robot.heightmap_resolution)
color_heightmap_image = Image.fromarray(color_heightmap)
depth_heightmap_image = Image.fromarray((depth_heightmap/0.1*255).astype(np.uint8))
for rotate_index in range(16):
rotated_color_heightmap = np.array(color_heightmap_image.rotate(angle = rotate_index*22.5, fillcolor = (0,0,0))).astype(np.uint8)
rotated_depth_heightmap = np.array(depth_heightmap_image.rotate(angle = rotate_index*22.5, fillcolor = 0)).astype(np.uint8)
color_depth_heightmap = np.concatenate((rotated_color_heightmap[:,:,[2,1,0]],rotated_depth_heightmap[:, :, np.newaxis]), axis = 2)[np.newaxis, :, :, :]
color_depth_heightmap = torch.from_numpy(color_depth_heightmap/255.0).permute(0,3,1,2).cuda().float()
with torch.no_grad():
net.eval()
predicted_score = net(color_depth_heightmap).permute(0,2,3,1)
soft_max_function = nn.Softmax(dim=3)
predicted_score = soft_max_function(predicted_score)[:,:,:,1]
predicted_score = torch.reshape(predicted_score, (predicted_score.shape[1], predicted_score.shape[2])).cpu().numpy()
torch.cuda.empty_cache()
max_score_pixel = np.unravel_index(np.argmax(predicted_score), predicted_score.shape)
max_score_pixel = [max_score_pixel[1], max_score_pixel[0]]
max_score_pixel_raw = [max_score_pixel]
max_score_pixel = point_position_after_rotation([max_score_pixel[0], htmap_h-max_score_pixel[1]], [htmap_w/2, htmap_h/2], -rotate_index*22.5) # plus or minus
max_score_pixel = [int(max_score_pixel[0]), int(htmap_h-max_score_pixel[1])]
if max_score_pixel[0]>=htmap_w or max_score_pixel[1]>=htmap_h or max_score_pixel[0]<0 or max_score_pixel[1]<0:
continue
pos = from_pixel_to_world_position(max_score_pixel, workspace_limits, heightmap_resolution, depth_heightmap)
yaw = from_angle_index_to_gripper_rot_z(rotate_index)
#pos = [pos[0]+0.002*sin(yaw), pos[1]+0.002*cos(yaw), pos[2]]
#print(np.max(predicted_score))
if np.max(predicted_score)>0.8:
thumb_position_predict = net_submodule2(color_depth_heightmap, torch.from_numpy(np.array(max_score_pixel_raw)).cuda().float()).cpu().detach().numpy().tolist()
if thumb_position_predict[0][0]<5:
continue
aperture_pixel = abs(thumb_position_predict-max_score_pixel_raw[0][1])[0][0]
#print(aperture_pixel)
aperture_real = aperture_pixel*heightmap_resolution*sin(60*pi/180)
max_score_pixel_list.append([pos, yaw, np.max(predicted_score),aperture_real])
max_score_pixel_list.sort(key=lambda x: x[2], reverse=True)
for max_score_pixel in max_score_pixel_list:
pos = max_score_pixel[0]
if pos[2]==0.03:
continue
yaw = max_score_pixel[1]
ini_aperture = max_score_pixel[3] #Go stone 0
ini_aperture = round(ini_aperture/0.005)*0.005
#color_heightmap_raw, _ = heightmap.get_heightmap(color_img_array, depth_array, cam_intrinsics, baseTcam, workspace_limits_raw, heightmap_resolution)
color_heightmap_raw, depth_heightmap_raw = heightmap.get_heightmap(color_img_array, depth_array, robot.cam_intrinsics, robot.baseTcam, workspace_limits_raw, robot.heightmap_resolution)
copy_color_heightmap = color_heightmap_raw.copy()
pix_x = int(max(min(round((pos[0] - workspace_limits_raw[0][0])/heightmap_resolution), 400-1), 0))
pix_y = int(max(min(round((pos[1] - workspace_limits_raw[1][0])/heightmap_resolution), 400-1), 0))
#cv2.circle(copy_color_heightmap, (pix_x,pix_y), 5, (0, 0, 255), 3)
pix_x_prime = int(max(min(round((pos[0]-ini_aperture*sin(yaw) - workspace_limits_raw[0][0])/heightmap_resolution), 400-1), 0))
pix_y_prime = int(max(min(round((pos[1]-ini_aperture*cos(yaw) - workspace_limits_raw[1][0])/heightmap_resolution), 400-1), 0))
pix_x_prime_enlong = int(max(min(round((pos[0]-(ini_aperture+0.003)*sin(yaw) - workspace_limits_raw[0][0])/heightmap_resolution), 400-1), 0))
pix_y_prime_enlong = int(max(min(round((pos[1]-(ini_aperture+0.003)*cos(yaw) - workspace_limits_raw[1][0])/heightmap_resolution), 400-1), 0))
pix_x_mid = int((pix_x+pix_x_prime)/2)
pix_y_mid = int((pix_y+pix_y_prime)/2)
color_difference1 = color_heightmap_raw[pix_y_mid, pix_x_mid]-color_heightmap_raw[pix_y_prime, pix_x_prime]
color_difference2 = color_heightmap_raw[pix_y_mid, pix_x_mid]-color_heightmap_raw[pix_y_prime_enlong, pix_x_prime_enlong]
from_thumb_to_finger_normal = (np.array([pix_x-pix_x_prime, pix_y-pix_y_prime])/np.linalg.norm(np.array([pix_x-pix_x_prime, pix_y-pix_y_prime]))).tolist()
perp_from_thumb_to_finger_normal = [-from_thumb_to_finger_normal[1], from_thumb_to_finger_normal[0]]
mid_a_little_left = [int(max(min(pix_x_mid+10*perp_from_thumb_to_finger_normal[0], 400-1), 0)), int(max(min(pix_y_mid+10*perp_from_thumb_to_finger_normal[1], 400-1), 0))]
mid_a_little_right = [int(max(min(pix_x_mid-10*perp_from_thumb_to_finger_normal[0], 400-1), 0)), int(max(min(pix_y_mid-10*perp_from_thumb_to_finger_normal[1], 400-1), 0))]
mid_a_little_left_depth = depth_heightmap_raw[mid_a_little_left[1], mid_a_little_left[0]]
mid_a_little_right_depth = depth_heightmap_raw[mid_a_little_right[1], mid_a_little_right[0]]
#print(mid_a_little_left, mid_a_little_right, mid_a_little_left_depth, mid_a_little_right_depth)
mid_point = [int(pix_x_mid[0]), int(pix_y_mid[1])]
if mid_point[1]-100<0:
subimage_top = 0
subimage_down = 200
elif mid_point[1]+100>400:
subimage_top = 200
subimage_down = 400
else:
subimage_top = mid_point[1]-100
subimage_down = mid_point[1]+100
if mid_point[0]-100<0:
subimage_left = 0
subimage_right = 200
elif mid_point[0]+100>400:
subimage_left = 200
subimage_right = 400
else:
subimage_left = mid_point[0]-100
subimage_right = mid_point[0]+100
sub_rgbd_image = rgbd_image[subimage_top: subimage_down, subimage_left: subimage_right][np.newaxis, :, :, :]
input_channel_sub_rgbd_image = np.concatenate(([sub_rgbd_image]*9), axis=0)
finger_position_in_image = (np.array([pix_x, pix_y]) - np.array([subimage_left, subimage_top])).tolist()
thumb_position_in_image = (np.array([pix_x_prime, pix_y_prime]) - np.array([subimage_left, subimage_top])).tolist()
input_channel_vector_list = []
for pitch in [60, 53, 45]:
for roll in [-10, 0, 10]:
input_channel_vector_list.append(finger_position_in_image+thumb_position_in_image+[pitch]+[roll])
input_channel_vector = np.array(input_channel_vector_list)
with torch.no_grad():
net_pitch_roll.eval()
predicted_score_pitch_roll = np.ravel(net(input_channel_sub_rgbd_image, input_channel_vector).cpu().numpy())
torch.cuda.empty_cache()
optimal_pitch_roll_index = np.argmax(predicted_score_pitch_roll)
if int(int(optimal_pitch_roll_index)/3)==0:
pitch = 60
elif int(int(optimal_pitch_roll_index)/3)==1:
pitch = 53
elif int(int(optimal_pitch_roll_index)/3)==2:
pitch = 45
if int(int(optimal_pitch_roll_index)%3)==0:
roll = -10
elif int(int(optimal_pitch_roll_index)%3)==1:
roll = 0
elif int(int(optimal_pitch_roll_index)%3)==2:
roll = 10
print('pos: ', pos, ' yaw: ', yaw, ' aperture: ', ini_aperture, 'pitch: ', pitch, ' roll: ', roll)
if robot.collision_check_scooping(pos, yaw, ini_aperture, theta=pitch*pi/180, roll=roll*pi/180)==True:
continue
cv2.arrowedLine(copy_color_heightmap, (pix_x_prime, pix_y_prime), (pix_x, pix_y), (0, 0, 255), 2)
test_image = cv2.flip(copy_color_heightmap, 1)
f = plt.figure(1)
plt.imshow(test_image[:,:,[2,1,0]])
plt.rcParams["keymap.quit"] = 'enter'
plt.show()
whether_collision = input("Whether collision? (y or n)")
if whether_collision=='y':
continue
if yaw>0:
pos=[pos[0]+0.001*sin(yaw), pos[1]+0.001*cos(yaw), pos[2]]
pass
grasp_success = robot.exe_scoop(pos, yaw, ini_aperture, thumb_extend=0.003, theta = pitch*pi/180, roll = roll*pi/180) #key 0.015 #Acrylic 0.008 #Go stone 0.003
total_index += 1
break
if __name__ == '__main__':
main()