-
Notifications
You must be signed in to change notification settings - Fork 2
/
superval_3dval.py
225 lines (183 loc) · 8.35 KB
/
superval_3dval.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
#!/usr/bin/python
import cv2
import time
import csv
import os
import sys
import rospy
import itertools
import numpy as np
#from tf.transformations import euler_from_quaternion
from cv_bridge import CvBridge
from namedlist import namedlist
from std_msgs.msg import Int64, String
from sensor_msgs.msg import CompressedImage, Image, JointState
from geometry_msgs.msg import Twist, Pose, TwistStamped, PoseStamped, Vector3
import torch
from model import Model
import argparse
import copy
from matplotlib import pyplot as plt
from blob import blob
class ImitateEval:
def __init__(self, weights):
self.bridge = CvBridge()
self.Data = namedlist('Data', ['pose', 'rgb', 'depth'])
self.data = self.Data(pose=None, rgb=None, depth=None)
self.is_start = True
checkpoint = torch.load(weights, map_location="cpu")
self.model = Model(**checkpoint['kwargs'])
self.model.load_state_dict(checkpoint["model_state_dict"])
self.model.eval()
def change_start(self):
radius = 0.07
# Publisher for the movement and the starting pose
self.movement_publisher = rospy.Publisher('/iiwa/CollisionAwareMotion', Pose, queue_size=10)
self.target_start = Pose()
self.target_start.position.x = -0.15 + np.random.rand()*2*radius - radius # -0.10757
self.target_start.position.y = 0.455 + np.random.rand()*2*radius - radius # 0.4103
self.target_start.position.z = 1.015
self.target_start.orientation.x = 0.0
self.target_start.orientation.y = 0.0
self.target_start.orientation.z = 0.7071068
self.target_start.orientation.w = 0.7071068
def move_to_button(self, tau, tolerance):
self.init_listeners()
rospy.Rate(5).sleep()
rate = rospy.Rate(15)
pose_to_move = copy.deepcopy(self.target_start)
eof = []
while not rospy.is_shutdown():
if None not in self.data:
# Position from the CartesianPose Topic!!
pos = self.data.pose.position
pos = [pos.x, pos.y, pos.z]
if self.is_start:
for _ in range(5):
eof += pos
self.is_start = False
else:
eof = pos + eof[:-3]
eof_input = torch.from_numpy(np.array(eof)).type(torch.FloatTensor)
eof_input = eof_input.unsqueeze(0)
rgb = self.process_images(self.data.rgb, True)
depth = self.process_images(self.data.depth, False)
# print("RGB min: {}, RGB max: {}".format(np.amin(rgb), np.amax(rgb)))
# print("Depth min: {}, Depth max: {}".format(np.amin(depth), np.amax(depth)))
print("EOF: {}".format(eof_input))
print("Tau: {}".format(tau))
torch.save(rgb, "/home/amazon/Desktop/rgb_tensor.pt")
torch.save(depth, "/home/amazon/Desktop/depth_tensor.pt")
torch.save(eof, "/home/amazon/Desktop/eof_tensor.pt")
torch.save(tau, "/home/amazon/Desktop/tau.pt")
with torch.no_grad():
out, aux = self.model(rgb, depth, eof_input, tau)
torch.save(out, "/home/amazon/Desktop/out.pt")
torch.save(aux, "/home/amazon/Desktop/aux.pt")
out = out.squeeze()
x_cartesian = out[0].item()
y_cartesian = out[1].item()
z_cartesian = out[2].item()
print("X:{}, Y:{}, Z:{}".format(x_cartesian, y_cartesian, z_cartesian))
print("Aux: {}".format(aux))
# This new pose is the previous pose + the deltas output by the net, adjusted for discrepancy in frame
# It used to be:
# pose_to_move.position.x += -y_cartesian
# pose_to_move.position.y += x_cartesian
# pose_to_move.position.z += z_cartesian
pose_to_move.position.x -= y_cartesian
pose_to_move.position.y += x_cartesian
pose_to_move.position.z += z_cartesian
#print(pose_to_move)
# Publish to Kuka!!!!
for i in range(10):
self.movement_publisher.publish(pose_to_move)
rospy.Rate(10).sleep()
rospy.wait_for_message("/iiwa/CollisionAwareExecutionStatus", String)
# End publisher
self.data = self.Data(pose=None,rgb=None,depth=None)
rate.sleep()
if distance(tau, (pose_to_move.position.x, pose_to_move.position.y, pose_to_move.position.z)) < tolerance:
break
def process_images(self, img_msg, is_it_rgb):
crop_right=586
crop_lower=386
img = self.bridge.compressed_imgmsg_to_cv2(img_msg, desired_encoding="passthrough")
if(is_it_rgb):
img = img[:,:,::-1]
# Does this crop work?
#rgb = img[0:386, 0:586]
#rgb = img.crop((0, 0, crop_right, crop_lower))
rgb = cv2.resize(img, (160,120))
rgb = np.array(rgb).astype(np.float32)
rgb = 2*((rgb - np.amin(rgb))/(np.amax(rgb)-np.amin(rgb)))-1
rgb = torch.from_numpy(rgb).type(torch.FloatTensor)
if is_it_rgb:
rgb = rgb.view(1, rgb.shape[0], rgb.shape[1], rgb.shape[2]).permute(0, 3, 1, 2)
else:
rgb = rgb.view(1, 1, rgb.shape[0], rgb.shape[1])
#plt.imshow(rgb[0,0] / 2 + .5)
# plt.show()
return rgb
def move_to_start(self):
# Publish starting position to Kuka!!!!
for i in range(10):
self.movement_publisher.publish(self.target_start)
rospy.Rate(10).sleep()
rospy.wait_for_message("/iiwa/CollisionAwareExecutionStatus", String)
# End publisher
def init_listeners(self):
# The Topics we are Subscribing to for data
self.right_arm_pose = rospy.Subscriber('/iiwa/state/CartesianPose', PoseStamped, self.pose_callback)
self.rgb_state_sub = rospy.Subscriber('/camera3/camera/color/image_rect_color/compressed', CompressedImage, self.rgb_callback)
self.depth_state_sub = rospy.Subscriber('/camera3/camera/depth/image_rect_raw/compressed', CompressedImage, self.depth_callback)
def unsubscribe(self):
self.right_arm_pose.unregister()
self.rgb_state_sub.unregister()
self.depth_state_sub.unregister()
def pose_callback(self, pose):
if None in self.data:
self.data.pose = pose.pose
def rgb_callback(self, rgb):
if None in self.data:
self.data.rgb = rgb
def depth_callback(self, depth):
if None in self.data:
self.data.depth = depth
def translate_tau(button):
b_0 = int(button[0])
b_1 = int(button[1])
tau = (-.22+.07*b_0, .56-.07*b_1, .94-.0025*b_y)
return tau
def distance(a, b):
return np.sqrt(np.sum([np.abs(aa - bb) for aa, bb in zip(a,b)]))
def get_tau(r, c):
tau = translate_tau([r, c])
def main(config):
for weights in list(blob(config.weights + '/*/best_checkpoint.tar', recursive=True)):
agent = Agent(config.weights)
agent.change_start()
agent.move_to_start()
rates = torch.zeros(3, 3, config.num_traj)
for r in range(3):
for c in range(3):
for i in range(config.num_traj):
tau = get_tau(r, c)
rates[3, c, i] = agent.move_to_button(tau, config.tolerance)
agent.change_start()
agent.move_to_start()
rates = torch.sum(rates, dim=2) / config.num_traj
torch.save(rates, config.weights[:config.weights.rfind('/')] + '/button_eval_percentages.pt')
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Arguments for evaluating imitation net")
parser.add_argument('-w', '--weights', required=True, help='Path to folder containing checkpoint directories/files.')
parser.add_argument('-t', '--tolerance', default=.01, type=float, help='Tolerance for button presses.')
parser.add_argument('-n', '--num_traj', default=10, type=int, help='Presses per button per arrangement.')
args = parser.parse_args()
rospy.init_node('eval_imitation', log_level=rospy.DEBUG)
try:
main(args)
except KeyboardInterrupt:
pass
cept KeyboardInterrupt:
pass