-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrf_pendulum_gpt4omini
199 lines (150 loc) · 7.3 KB
/
rf_pendulum_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
Generation 0
import numpy as np
def reward(action, obs):
"""
Calculates the reward for the pendulum based on its angle and angular velocity.
Parameters:
action (np.ndarray): An array representing the torque applied to the pendulum, shape (1,).
obs (np.ndarray): An array representing the current state of the pendulum, shape (2,),
where obs[0] is the angle (0 to 2*pi) and obs[1] is the angular velocity
(-12*pi to 12*pi).
Returns:
float: The calculated reward based on the pendulum's state.
"""
# Extracting angle and angular velocity from the observation
angle = obs[0]
angular_velocity = obs[1]
# Normalize angle difference to be within [-pi, pi]
angle_diff = np.abs((angle - np.pi + np.pi) % (2 * np.pi) - np.pi)
# Calculate rewards based on angle difference with a potential threshold
if angle_diff < 0.1: # 10 degrees tolerance
rew_angle = 1.0 # Full reward close to upright
elif angle_diff < 0.5: # 30 degrees tolerance
rew_angle = 1.0 - (angle_diff / 0.5) # Linear decay
else:
rew_angle = 0.0 # No reward if too far from upright
# Penalizing for high angular velocity with a sharper penalty function
rew_velocity = np.exp(-np.clip(np.abs(angular_velocity) / 3, 0, 5)) # reduce scale
# Combined reward with normalization to mitigate large action effects
rew = rew_angle * rew_velocity
return rew
Generation 1
import numpy as np
def reward(action, obs):
"""
Calculates the reward for the pendulum based on its angle and angular velocity.
Parameters:
action (np.ndarray): An array representing the torque applied to the pendulum, shape (1,).
obs (np.ndarray): An array representing the current state of the pendulum, shape (2,),
where obs[0] is the angle (0 to 2*pi) and obs[1] is the angular velocity
(-12*pi to 12*pi).
Returns:
float: The calculated reward based on the pendulum's state.
"""
# Extracting angle and angular velocity from the observation
angle = obs[0]
angular_velocity = obs[1]
# Normalize angle difference to be within [-pi, pi]
angle_diff = np.abs((angle - np.pi + np.pi) % (2 * np.pi) - np.pi)
# Calculate rewards based on angle difference with a potential threshold
if angle_diff < 0.05: # 5 degrees tolerance
rew_angle = 1.0 # Full reward close to upright
elif angle_diff < 0.3: # 17 degrees tolerance
rew_angle = 1.0 - (angle_diff / 0.3) # Linear decay
else:
rew_angle = 0.0 # No reward if too far from upright
# Penalizing for high angular velocity with a sharper penalty function
rew_velocity = 1.0 - np.clip(np.abs(angular_velocity) / 12, 0, 1) # Scaling the penalty
# Combined reward with emphasis on angle, ensuring high values for both metrics
rew = rew_angle * rew_velocity
return rew
Generation 2
import numpy as np
def reward(action, obs):
"""
Calculates the reward for the pendulum based on its angle and angular velocity.
Parameters:
action (np.ndarray): An array representing the torque applied to the pendulum, shape (1,).
obs (np.ndarray): An array representing the current state of the pendulum, shape (2,),
where obs[0] is the angle (0 to 2*pi) and obs[1] is the angular velocity
(-12*pi to 12*pi).
Returns:
float: The calculated reward based on the pendulum's state.
"""
# Extracting angle and angular velocity from the observation
angle = obs[0]
angular_velocity = obs[1]
# Normalize angle difference to be within [-pi, pi]
angle_diff = np.abs((angle - np.pi + np.pi) % (2 * np.pi) - np.pi)
# Calculate rewards based on angle difference with a potential threshold
if angle_diff < 0.05: # 5 degrees tolerance
rew_angle = 1.0 # Full reward close to upright
elif angle_diff < 0.3: # 17 degrees tolerance
rew_angle = 1.0 - (angle_diff / 0.3) # Linear decay
else:
rew_angle = 0.0 # No reward if too far from upright
# Penalizing for high angular velocity with a sharper penalty function
rew_velocity = 1.0 - np.clip(np.abs(angular_velocity) / 12, 0, 1) # Scaling the penalty
# Combined reward with emphasis on angle, ensuring high values for both metrics
rew = rew_angle * rew_velocity
return rew
Generation 3
import numpy as np
def reward(action, obs):
"""
Calculates the reward for the pendulum based on its angle and angular velocity.
Parameters:
action (np.ndarray): An array representing the torque applied to the pendulum, shape (1,).
obs (np.ndarray): An array representing the current state of the pendulum, shape (2,),
where obs[0] is the angle (0 to 2*pi) and obs[1] is the angular velocity
(-12*pi to 12*pi).
Returns:
float: The calculated reward based on the pendulum's state.
"""
# Extracting angle and angular velocity from the observation
angle = obs[0]
angular_velocity = obs[1]
# Normalize angle difference to be within [-pi, pi]
angle_diff = np.abs((angle - np.pi + np.pi) % (2 * np.pi) - np.pi)
# Calculate rewards based on angle difference with a potential threshold
if angle_diff < 0.05: # 5 degrees tolerance
rew_angle = 1.0 # Full reward close to upright
elif angle_diff < 0.3: # 17 degrees tolerance
rew_angle = 1.0 - (angle_diff / 0.3) # Linear decay
else:
rew_angle = 0.0 # No reward if too far from upright
# Penalizing for high angular velocity with a sharper penalty function
rew_velocity = 1.0 - np.clip(np.abs(angular_velocity) / 12, 0, 1) # Scaling the penalty
# Combined reward with emphasis on angle, ensuring high values for both metrics
rew = rew_angle * rew_velocity
return rew
Generation 4
import numpy as np
def reward(action, obs):
"""
Calculates the reward for the pendulum based on its angle and angular velocity.
Parameters:
action (np.ndarray): An array representing the torque applied to the pendulum, shape (1,).
obs (np.ndarray): An array representing the current state of the pendulum, shape (2,),
where obs[0] is the angle (0 to 2*pi) and obs[1] is the angular velocity
(-12*pi to 12*pi).
Returns:
float: The calculated reward based on the pendulum's state.
"""
# Extracting angle and angular velocity from the observation
angle = obs[0]
angular_velocity = obs[1]
# Normalize angle difference to be within [-pi, pi]
angle_diff = np.abs((angle - np.pi + np.pi) % (2 * np.pi) - np.pi)
# Calculate rewards based on angle difference with a potential threshold
if angle_diff < 0.05: # 5 degrees tolerance
rew_angle = 1.0 # Full reward close to upright
elif angle_diff < 0.3: # 17 degrees tolerance
rew_angle = 1.0 - (angle_diff / 0.3) # Linear decay
else:
rew_angle = 0.0 # No reward if too far from upright
# Penalizing for high angular velocity with a sharper penalty function
rew_velocity = 1.0 - np.clip(np.abs(angular_velocity) / 12, 0, 1) # Scaling the penalty
# Combined reward with emphasis on angle, ensuring high values for both metrics
rew = rew_angle * rew_velocity
return rew