forked from praveenVnktsh/soundsense
-
Notifications
You must be signed in to change notification settings - Fork 0
/
robot_node_vis.py
209 lines (179 loc) · 6.82 KB
/
robot_node_vis.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
import numpy as np
import torch
import stretch_body.robot
import time
import cv2
import yaml
import rospy
from audio_common_msgs.msg import AudioDataStamped, AudioData
class RobotNode:
def __init__(self, config_path, model, is_unimodal = False):
self.r = stretch_body.robot.Robot()
self.boot_robot()
with open(config_path) as info:
params = yaml.load(info.read(), Loader=yaml.FullLoader)
self.image_shape = params['camera_inp_h_w']
self.bgr_to_rgb = params['bgr_to_rgb']
self.hz = params['audio_hz']
self.audio_n_seconds = params['audio_history_seconds']
cam = params['camera_id']
self.n_stack_images = params['camera_stack_images']
self.history = {
'audio': [],
'video': [],
}
if not is_unimodal:
audio_sub = rospy.Subscriber('/audio/audio', AudioData, self.callback)
print("Waiting for audio data...")
rospy.wait_for_message('/audio/audio', AudioData, timeout=10)
self.cap = cv2.VideoCapture(cam)
self.model = model
def callback(self, data):
audio = np.frombuffer(data.data, dtype=np.uint8)
self.history['audio'].append(audio)
if len(self.history['audio']) > self.audio_n_seconds * self.hz:
self.history['audio'] = self.history['audio'][-self.audio_n_seconds * self.hz:]
def boot_robot(self,):
r = self.r
if not r.startup():
print("Failed to start robot")
exit() # failed to start robot!
if not r.is_homed():
print("Robot is not calibrated. Do you wish to calibrate? (y/n)")
if input() == "y":
r.home()
else:
print("Exiting...")
exit()
r.lift.move_to(0.9)
r.arm.move_to(0.3)
r.end_of_arm.move_to('wrist_yaw', 0.0)
r.end_of_arm.move_to('wrist_pitch', 0.0)
r.end_of_arm.move_to('wrist_roll', 0.0)
r.end_of_arm.move_to('stretch_gripper', 100)
r.push_command()
r.lift.wait_until_at_setpoint()
r.arm.wait_until_at_setpoint()
time.sleep(5)
print("Robot ready to run model.")
def get_image(self):
h, w = self.image_shape
ret, frame = self.cap.read()
frame = cv2.resize(frame, (h, w))
if self.bgr_to_rgb:
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if not ret:
return None
return frame
def run_loop(self, visualize = False):
is_run = True
start_time = time.time()
loop_rate = 10
loop_start_time = time.time()
n_stack = self.n_stack_images * self.audio_n_seconds
attention_scores = np.zeros(shape=(1,8,1,1))
while is_run:
frame = self.get_image()
if len(self.history['video']) > 0:
self.history['video'][-1] = frame.copy()
if time.time() - loop_start_time > 1/loop_rate:
loop_start_time = time.time()
self.history['video'].append(frame)
if frame is not None:
if len(self.history['video']) < n_stack:
continue
if len(self.history['video']) > n_stack:
self.history['video'] = self.history['video'][-n_stack:]
# if visualize:
# stacked_inp = np.hstack(self.history['video'])
# cv2.imwrite('stacked_inp.jpg', stacked_inp)
# cv2.imshow('stacked', stacked_inp)
# cv2.waitKey(1)
attention_scores = self.execute_action(attention_scores)
# Save attention scores
with open('your_file.txt', 'w') as f:
for line in attention_scores:
f.write(f"{line}\n")
else:
print("No frame")
is_run = False
np.save('attention_scores.npy', attention_scores)
print("Ending run loop")
def generate_inputs(self, save = True):
video = self.history['video'] # subsample n_frames of interest
n_images = len(video)
choose_every = n_images // self.n_stack_images
video = video[::choose_every]
audio = self.history['audio']
if save:
stacked = np.hstack(video)
cv2.imwrite('stacked.jpg', stacked)
# cam_gripper_framestack,audio_clip_g
# vg_inp: [batch, num_stack, 3, H, W]
# a_inp: [batch, 1, T]
video = video[-self.n_stack_images:]
video = [(img).astype(float)/ 255 for img in video]
return {
'video' : video, # list of images
'audio' : audio, # audio buffer
}
def execute_action(self, attention_scores):
# lift, extension, lateral, roll, gripper
# action_space = [del_extension, del_height, del_lateral, del_roll, del_gripper]
# action[0] = np.clip(action[0], -0.01, 0.01)
r = self.r
inputs = self.generate_inputs()
outputs, weights = self.model(inputs) # 11 dimensional
attention_scores += weights.detch().numpy()
# w - extend arm
# s - retract arm
# a - move left
# d - move right
# i - lift up
# k - lift down
# l - roll right
# j - roll left
# m - close gripper
# n - open gripper
mapping = {
'w': 0,
's': 1,
'a': 2,
'd': 3,
'n': 4,
'm': 5,
'i': 6,
'k': 7,
'j': 8,
'l': 9,
'none': 10,
}
inv_mapping = {v: k for k, v in mapping.items()}
action_idx = torch.argmax(outputs).item()
print("action: ", inv_mapping[action_idx], "output: ", outputs)
big = 0.05
movement_resolution = {
0: big,
1: -big,
2: big,
3: -big,
4: 50,
5: -50,
6: big,
7: -big,
8: 15 * np.pi/180,
9: -15 * np.pi/180,
}
if action_idx in [0, 1]:
r.arm.move_by(movement_resolution[action_idx])
elif action_idx in [2, 3]:
r.base.translate_by(movement_resolution[action_idx])
elif action_idx in [4, 5]:
r.end_of_arm.move_by('stretch_gripper', movement_resolution[action_idx])
elif action_idx in [8, 9]:
r.end_of_arm.move_by('wrist_roll', movement_resolution[action_idx])
elif action_idx in [6, 7]:
r.lift.move_by(movement_resolution[action_idx])
r.push_command()
# time.sleep(1)
return attention_scores