This repository has been archived by the owner on Nov 19, 2023. It is now read-only.
/
kinova.py
209 lines (158 loc) · 6.96 KB
/
kinova.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 sys
import os
import time
import threading
import time
from kortex_api.autogen.client_stubs.BaseClientRpc import BaseClient
from kortex_api.autogen.client_stubs.BaseCyclicClientRpc import BaseCyclicClient
from kortex_api.autogen.messages import Base_pb2, BaseCyclic_pb2, Common_pb2
import utilities
import gym
import numpy as np
from gym.envs.registration import register
import cv2
def register_env():
register(
id='KinovaReach-v0',
entry_point='kinova:KinovaRobotEnv',
)
class KinovaRobotEnv:
metadata = {'render_modes':['human', 'rbg_array']}
def __init__(self, action_scale=0.05, target_range=0.15, distance_threshold=0.05, speed=0.10):
self.connection = utilities.DeviceConnection.createTcpConnection()
self.router = self.connection.__enter__()
self.base = BaseClient(self.router)
self.base_cyclic = BaseCyclicClient(self.router)
self.webcam = cv2.VideoCapture(0)
self.action_scale = action_scale
self.target_range = target_range
self.distance_threshold = distance_threshold
self.speed = speed
self.observation_space = gym.spaces.Dict(
observation=gym.spaces.Box(-np.inf, np.inf, self._get_obs().shape),
achieved_goal=gym.spaces.Box(-np.inf, np.inf, (3,)),
desired_goal=gym.spaces.Box(-np.inf, np.inf, (3,)),
)
self.action_space = gym.spaces.Box(-1., 1., (3,))
self.reward_range = (-np.inf, 0)
self.unwrapped = self
self.spec = None
self._set_to_home()
self.init_xyz = self._get_obs()[:3]
self.goal = None
self.sample_goal()
assert self.goal is not None
def seed(self, seed):
pass
def close(self):
self.connection.__exit__(None, None, None)
def render(self, mode='human', width=84, height=84):
# return np.zeros((width, height, 3))
ret, frame = self.webcam.read()
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if mode == 'human':
cv2.imshow('', frame)
elif mode == 'rgb_array':
frame = cv2.resize(frame, (width, height))
return frame
def sample_goal(self, target_range=None):
if target_range is None:
target_range = self.target_range
self.goal = self.init_xyz + (-1 + 2*np.random.rand(*self.init_xyz.shape))*self.target_range
return self.goal
def reset(self):
self.sample_goal()
self._set_to_home()
return self._get_obs_dict()
def step(self, act):
print(f'act: {act}')
act = act.copy()*self.action_scale
action = Base_pb2.Action()
action.name = "Example Cartesian action movement"
action.application_data = ""
action.reach_pose.constraint.speed.translation = self.speed
feedback = self.base_cyclic.RefreshFeedback()
cartesian_pose = action.reach_pose.target_pose
cartesian_pose.x = feedback.base.tool_pose_x + act[0] # (meters)
cartesian_pose.y = feedback.base.tool_pose_y + act[1] # (meters)
cartesian_pose.z = feedback.base.tool_pose_z + act[2] # (meters)
cartesian_pose.theta_x = feedback.base.tool_pose_theta_x # (degrees)
cartesian_pose.theta_y = feedback.base.tool_pose_theta_y # (degrees)
cartesian_pose.theta_z = feedback.base.tool_pose_theta_z # (degrees)
self.base.ExecuteAction(action)
duration = 1. + (np.linalg.norm(act)/self.speed)
print(f'step w duration {duration}')
time.sleep(duration)
obs_dict = self._get_obs_dict()
info = dict()
reward = self.compute_reward(
obs_dict['achieved_goal'],
obs_dict['desired_goal'],
info,
)
done = True if np.abs(reward) < self.distance_threshold else False
info['is_success'] = done
return obs_dict, reward, done, info
def _set_to(self, xyz):
action = Base_pb2.Action()
action.name = "Example Cartesian action movement"
action.application_data = ""
action.reach_pose.constraint.speed.translation = self.speed
feedback = self.base_cyclic.RefreshFeedback()
current_pos = np.asarray([feedback.base.tool_pose_x, feedback.base.tool_pose_y, feedback.base.tool_pose_z])
distance_to_xyz = np.linalg.norm(current_pos - xyz)
duration = 1. + (distance_to_xyz/self.speed)
print(f'_set_to w duration {duration}')
cartesian_pose = action.reach_pose.target_pose
cartesian_pose.x = xyz[0]
cartesian_pose.y = xyz[1]
cartesian_pose.z = xyz[2]
cartesian_pose.theta_x = feedback.base.tool_pose_theta_x # (degrees)
cartesian_pose.theta_y = feedback.base.tool_pose_theta_y # (degrees)
cartesian_pose.theta_z = feedback.base.tool_pose_theta_z # (degrees)
self.base.ExecuteAction(action)
time.sleep(duration)
def _set_to_home(self):
# Make sure the arm is in Single Level Servoing mode
base_servo_mode = Base_pb2.ServoingModeInformation()
base_servo_mode.servoing_mode = Base_pb2.SINGLE_LEVEL_SERVOING
self.base.SetServoingMode(base_servo_mode)
# Move arm to ready position
print("Moving the arm to a safe position")
action_type = Base_pb2.RequestedActionType()
action_type.action_type = Base_pb2.REACH_JOINT_ANGLES
action_list = self.base.ReadAllActions(action_type)
action_handle = None
for action in action_list.action_list:
if action.name == "Home":
action_handle = action.handle
if action_handle == None:
raise Run("Can't reach safe position. Exiting")
self.base.ExecuteActionFromReference(action_handle)
print('Resetting for duration 6')
time.sleep(6)
def compute_reward(self, achieved_goal, desired_goal, info):
d = np.linalg.norm(achieved_goal - desired_goal, axis=-1)
return 0. if d < self.distance_threshold else -1.
def _is_success(self, achieved_goal, desired_goal):
d = np.linalg.norm(achieved_goal - desired_goal, axis=-1)
return True if d < self.distance_threshold else False
def _get_obs(self):
feedback = self.base_cyclic.RefreshFeedback()
observation = np.asarray([
feedback.base.tool_pose_x,
feedback.base.tool_pose_y,
feedback.base.tool_pose_z,
])
return observation
def _get_obs_dict(self):
observation = self._get_obs()
achieved_goal = observation[:3]
desired_goal = self.goal
return dict(
observation=observation,
achieved_goal=achieved_goal,
desired_goal=desired_goal,
)
if __name__ == "__main__":
env = KinovaRobotEnv()