-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrf_flyer2d_gpt4omini
279 lines (212 loc) · 10.5 KB
/
rf_flyer2d_gpt4omini
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
Generation 0
import numpy as np
def reward(action, obs):
"""
Calculate the reward for the 2D Flyer system based on its current state.
Parameters:
action (np.ndarray): The forces applied by the left and right propellers, shape (2,).
obs (np.ndarray): The current state of the system containing position, velocity, and angle,
shape (6,).
Returns:
float: The calculated reward.
"""
# Desired goal state
goal_state = np.array([0, 0, 0, 0, 0, 0])
# Calculate the distance from the goal state
state_error = obs - goal_state
# Define a threshold for being "close enough" to the goal state
position_threshold = 0.05 # 5 cm
velocity_threshold = 0.5 # 0.5 m/s
angle_threshold = 0.1 # 0.1 rad
# Calculate the distance in position and velocity
position_error = np.linalg.norm(state_error[:2]) # Euclidean distance in 2D position space
angle_error = abs(state_error[2]) # Angle error
velocity_error = np.linalg.norm(state_error[3:5]) # Velocity in the 2D velocity space
# Initialize the reward
rew = 0.0
# Reward for being close to the goal state
if all(abs(state_error) < np.array([position_threshold, position_threshold, angle_threshold,
velocity_threshold, velocity_threshold, 10 * np.pi])):
rew += 100.0 # High reward for reaching or being close to the goal
# Penalize for being too far from the goal state
else:
# Calculate penalties for position, angle, and velocity
if position_error > position_threshold:
rew -= 10.0 * position_error # Linear penalty for position error
if angle_error > angle_threshold:
rew -= 5.0 * angle_error # Linear penalty for angle error
if velocity_error > velocity_threshold:
rew -= 2.0 * velocity_error # Linear penalty for velocity error
# Encourage lesser forces applied to maintain stability
force_penalty = np.linalg.norm(action) # Penalize based on the magnitude of the forces applied
rew -= 5.0 * force_penalty # Stabilizing penalty
return rew
Generation 1
import numpy as np
def reward(action, obs):
"""
Calculate the reward for the 2D Flyer system based on its current state.
Parameters:
action (np.ndarray): The forces applied by the left and right propellers, shape (2,).
obs (np.ndarray): The current state of the system containing position, velocity, and angle,
shape (6,).
Returns:
float: The calculated reward.
"""
# Desired goal state
goal_state = np.array([0, 0, 0, 0, 0, 0])
# Calculate the distance from the goal state
state_error = obs - goal_state
# Define a threshold for being "close enough" to the goal state
position_threshold = 0.05 # 5 cm
velocity_threshold = 0.5 # 0.5 m/s
angle_threshold = 0.1 # 0.1 rad
# Calculate the distance in position and velocity
position_error = np.linalg.norm(state_error[:2]) # Euclidean distance in 2D position space
angle_error = abs(state_error[2]) # Angle error
velocity_error = np.linalg.norm(state_error[3:5]) # Velocity in the 2D velocity space
# Initialize the reward
rew = 0.0
# Reward for being close to the goal state
if all(abs(state_error) < np.array([position_threshold, position_threshold, angle_threshold,
velocity_threshold, velocity_threshold, 10 * np.pi])):
rew += 100.0 # High reward for reaching or being close to the goal
# Penalize for being too far from the goal state
else:
# Calculate penalties for position, angle, and velocity
if position_error > position_threshold:
rew -= 10.0 * position_error # Linear penalty for position error
if angle_error > angle_threshold:
rew -= 5.0 * angle_error # Linear penalty for angle error
if velocity_error > velocity_threshold:
rew -= 2.0 * velocity_error # Linear penalty for velocity error
# Encourage lesser forces applied to maintain stability
force_penalty = np.linalg.norm(action) # Penalize based on the magnitude of the forces applied
rew -= 5.0 * force_penalty # Stabilizing penalty
return rew
Generation 2
import numpy as np
def reward(action, obs):
"""
Calculate the reward for the 2D Flyer system based on its current state.
Parameters:
action (np.ndarray): The forces applied by the left and right propellers, shape (2,).
obs (np.ndarray): The current state of the system containing position, velocity, and angle,
shape (6,).
Returns:
float: The calculated reward.
"""
# Desired goal state
goal_state = np.array([0, 0, 0, 0, 0, 0])
# Calculate the distance from the goal state
state_error = obs - goal_state
# Define a threshold for being "close enough" to the goal state
position_threshold = 0.05 # 5 cm
velocity_threshold = 0.5 # 0.5 m/s
angle_threshold = 0.1 # 0.1 rad
# Calculate the distance in position and velocity
position_error = np.linalg.norm(state_error[:2]) # Euclidean distance in 2D position space
angle_error = abs(state_error[2]) # Angle error
velocity_error = np.linalg.norm(state_error[3:5]) # Velocity in the 2D velocity space
# Initialize the reward
rew = 0.0
# Reward for being close to the goal state
if all(abs(state_error) < np.array([position_threshold, position_threshold, angle_threshold,
velocity_threshold, velocity_threshold, 10 * np.pi])):
rew += 100.0 # High reward for reaching or being close to the goal
# Penalize for being too far from the goal state
else:
# Calculate penalties for position, angle, and velocity
if position_error > position_threshold:
rew -= 10.0 * position_error # Linear penalty for position error
if angle_error > angle_threshold:
rew -= 5.0 * angle_error # Linear penalty for angle error
if velocity_error > velocity_threshold:
rew -= 2.0 * velocity_error # Linear penalty for velocity error
# Encourage lesser forces applied to maintain stability
force_penalty = np.linalg.norm(action) # Penalize based on the magnitude of the forces applied
rew -= 5.0 * force_penalty # Stabilizing penalty
return rew
Generation 3
import numpy as np
def reward(action, obs):
"""
Calculate the reward for the 2D Flyer system based on its current state.
Parameters:
action (np.ndarray): The forces applied by the left and right propellers, shape (2,).
obs (np.ndarray): The current state of the system containing position, velocity, and angle,
shape (6,).
Returns:
float: The calculated reward.
"""
# Desired goal state
goal_state = np.array([0, 0, 0, 0, 0, 0])
# Calculate the distance from the goal state
state_error = obs - goal_state
# Define a threshold for being "close enough" to the goal state
position_threshold = 0.05 # 5 cm
velocity_threshold = 0.5 # 0.5 m/s
angle_threshold = 0.1 # 0.1 rad
# Calculate the distance in position and velocity
position_error = np.linalg.norm(state_error[:2]) # Euclidean distance in 2D position space
angle_error = abs(state_error[2]) # Angle error
velocity_error = np.linalg.norm(state_error[3:5]) # Velocity in the 2D velocity space
# Initialize the reward
rew = 0.0
# Reward for being close to the goal state
if all(abs(state_error) < np.array([position_threshold, position_threshold, angle_threshold,
velocity_threshold, velocity_threshold, 10 * np.pi])):
rew += 100.0 # High reward for reaching or being close to the goal
# Penalize for being too far from the goal state
else:
# Calculate penalties for position, angle, and velocity
if position_error > position_threshold:
rew -= 10.0 * position_error # Linear penalty for position error
if angle_error > angle_threshold:
rew -= 5.0 * angle_error # Linear penalty for angle error
if velocity_error > velocity_threshold:
rew -= 2.0 * velocity_error # Linear penalty for velocity error
# Encourage lesser forces applied to maintain stability
force_penalty = np.linalg.norm(action) # Penalize based on the magnitude of the forces applied
rew -= 5.0 * force_penalty # Stabilizing penalty
return rew
Generation 4
import numpy as np
def reward(action, obs):
"""
Calculate the reward for the 2D Flyer system based on its current state.
Parameters:
action (np.ndarray): The forces applied by the left and right propellers, shape (2,).
obs (np.ndarray): The current state of the system containing position, velocity, and angle,
shape (6,).
Returns:
float: The calculated reward.
"""
# Desired goal state
goal_state = np.array([0, 0, 0, 0, 0, 0])
# Calculate the distance from the goal state
state_error = obs - goal_state
# Define a threshold for being "close enough" to the goal state
position_threshold = 0.05 # 5 cm
velocity_threshold = 0.5 # 0.5 m/s
angle_threshold = 0.1 # 0.1 rad
# Calculate the state errors
position_error = np.linalg.norm(state_error[:2]) # Euclidean distance in 2D position space
angle_error = abs(state_error[2]) # Angle error
velocity_error = np.linalg.norm(state_error[3:5]) # Velocity in the 2D velocity space
# Initialize the reward
rew = 0.0
# Reward structure based on proximity to goal state
if position_error < position_threshold and angle_error < angle_threshold and velocity_error < velocity_threshold:
rew += 100.0 # High reward for reaching or being close to the goal
else:
# Scaled penalties for position, angle, and velocity errors using exponential decay
rew -= 20.0 * position_error
rew -= 10.0 * angle_error
rew -= 5.0 * velocity_error
# Encourage stability by penalizing high force magnitudes
force_penalty = np.linalg.norm(action)
rew -= 5.0 * force_penalty # Penalize large forces used
# Add a small negative reward for every time step to encourage quicker convergence
rew -= 0.1
return rew