-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrf_flyer2d_gemma2
160 lines (105 loc) · 4.87 KB
/
rf_flyer2d_gemma2
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
Generation 0
import numpy as np
def reward(action, obs):
"""
Reward function for the 2D Flyer system.
Args:
action (np.ndarray): A 2-dimensional array representing the forces applied by the left and right propellers.
obs (np.ndarray): The current state of the system, a 6-dimensional array representing position, velocity, and angle.
Returns:
float: The reward for the given state and action.
"""
goal_state = np.zeros(6) # Desired goal state
position_error = np.linalg.norm(obs[:2] - goal_state[:2])
angle_error = np.abs(obs[2] ) # Angular error
velocity_penalty = 0.1 * np.linalg.norm(obs[3:]) # Penalize velocity
# Shaping reward to encourage achieving and staying in the goal state
reward = -0.1*position_error - angle_error - velocity_penalty
if position_error < 0.1 and angle_error < 0.5: # Within a small radius of the goal
reward += 100.0
elif position_error < 0.5 :
reward += 1.
return reward
Generation 1
import numpy as np
def reward(action, obs):
"""
Reward function for the 2D Flyer system.
Args:
action (np.ndarray): A 2-dimensional array representing the forces applied by the left and right propellers.
obs (np.ndarray): The current state of the system, a 6-dimensional array representing position, velocity, and angle.
Returns:
float: The reward for the given state and action.
"""
goal_state = np.zeros(6) # Desired goal state
position_error = np.linalg.norm(obs[:2] - goal_state[:2])
angle_error = np.abs(obs[2]) # Angular error normalized by pi
velocity_penalty = 0.1 * np.linalg.norm(obs[3:]) # Penalize velocity
# Shaping reward to encourage achieving and staying in the goal state
reward = -position_error - angle_error/np.pi- velocity_penalty
if position_error < 0.1 and angle_error < 0.5: # Within a small radius of the goal
reward += 100.0
elif position_error < 0.3 : # Relaxed reward to encourage staying near goal state
reward += 1.
return reward
Generation 2
import numpy as np
def reward(action, obs):
"""
Reward function for the 2D Flyer system.
Args:
action (np.ndarray): A 2-dimensional array representing the forces applied by the left and right propellers.
obs (np.ndarray): The current state of the system, a 6-dimensional array representing position, velocity, and angle.
Returns:
float: The reward for the given state and action.
"""
goal_state = np.zeros(6) # Desired goal state
position_error = np.linalg.norm(obs[:2] - goal_state[:2])
angle_error = obs[2]**2 # Penalize squared angle error for smoother control
velocity_penalty = 0.1 * np.linalg.norm(obs[3:]) # Penalize velocity
reward = -position_error - angle_error- velocity_penalty
if position_error < 0.1 and abs(angle_error) < 0.25: # Tighten goal threshold for better precision
reward += 100.0
return reward
Generation 3
import numpy as np
def reward(action, obs):
"""
Reward function for the 2D Flyer system.
Args:
action (np.ndarray): A 2-dimensional array representing the forces applied by the left and right propellers.
obs (np.ndarray): The current state of the system, a 6-dimensional array representing position, velocity, and angle.
Returns:
float: The reward for the given state and action.
"""
goal_state = np.zeros(6)
position_error = np.linalg.norm(obs[:2] - goal_state[:2])
angle_error = obs[2]**2
velocity_penalty = 0.1 * np.linalg.norm(obs[3:])
action_penalty = 0.01 * np.linalg.norm(action) # Penalize large actions for smoothness
reward = -position_error - angle_error - velocity_penalty - action_penalty
if position_error < 0.05 and abs(angle_error) < 0.1:
reward += 100.0
return reward
Generation 4
import numpy as np
def reward(action, obs):
"""
Reward function for the 2D Flyer system.
Args:
action (np.ndarray): A 2-dimensional array representing the forces applied by the left and right propellers.
obs (np.ndarray): The current state of the system, a 6-dimensional array representing position, velocity, and angle.
Returns:
float: The reward for the given state and action.
"""
goal_state = np.zeros(6)
position_error = np.linalg.norm(obs[:2] - goal_state[:2])
angle_error = obs[2]**2
velocity_penalty = 0.1 * np.linalg.norm(obs[3:])
action_penalty = 0.01 * np.linalg.norm(action) # Penalize large actions for smoothness
# Increase reward scaling for proximity to the goal
reward = -position_error - angle_error - velocity_penalty
reward -= action_penalty
if position_error < 0.05 and abs(angle_error) < 0.1:
reward += 5 * (1 - position_error) # More reward for being closer
return reward