-
Notifications
You must be signed in to change notification settings - Fork 1
/
cartpole_full.py
226 lines (195 loc) · 8.84 KB
/
cartpole_full.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
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
"""
Classic cart-pole system implemented by Rich Sutton et al.
Copied from http://incompleteideas.net/sutton/book/code/pole.c
permalink: https://perma.cc/C9ZM-652R
"""
import math
import gym
from gym import spaces, logger
from gym.utils import seeding
import numpy as np
import csv
class CartPoleEnv_adv(gym.Env):
"""
Description:
A pole is attached by an un-actuated joint to a cart, which moves along a frictionless track. The pendulum starts upright, and the goal is to prevent it from falling over by increasing and reducing the cart's velocity.
Source:
This environment corresponds to the version of the cart-pole problem described by Barto, Sutton, and Anderson
Observation:
Type: Box(4)
Num Observation Min Max
0 Cart Position -4.8 4.8
1 Cart Velocity -Inf Inf
2 Pole Angle -24° 24°
3 Pole Velocity At Tip -Inf Inf
Actions:
Type: Discrete(2)
Num Action
0 Push cart to the left
1 Push cart to the right
Note: The amount the velocity is reduced or increased is not fixed as it depends on the angle the pole is pointing. This is because the center of gravity of the pole increases the amount of energy needed to move the cart underneath it
Reward:
Reward is 1 for every step taken, including the termination step
Starting State:
All observations are assigned a uniform random value between ±0.05
Episode Termination:
Pole Angle is more than ±12°
Cart Position is more than ±2.4 (center of the cart reaches the edge of the display)
Episode length is greater than 200
Solved Requirements
Considered solved when the average reward is greater than or equal to 195.0 over 100 consecutive trials.
"""
metadata = {
'render.modes': ['human', 'rgb_array'],
'video.frames_per_second': 50
}
def __init__(self):
self.gravity = 10
# 1 0.1 0.5 original
self.masscart = 1
self.masspole = 0.1
self.total_mass = (self.masspole + self.masscart)
self.length = 0.5 # actually half the pole's length
self.polemass_length = (self.masspole * self.length)
self.force_mag = 20
self.tau = 0.02 # seconds between state updates
self.kinematics_integrator = 'euler'
# Angle at which to fail the episode
self.theta_threshold_radians = 20 * 2 * math.pi / 360
# self.theta_threshold_radians = 12 * 2 * math.pi / 360
self.x_threshold = 5
# self.max_v=1.5
# self.max_w=1
# FOR DATA
self.max_v = 50
self.max_w = 50
# Angle limit set to 2 * theta_threshold_radians so failing observation is still within bounds
high = np.array([
self.x_threshold * 2,
self.max_v,
self.theta_threshold_radians * 2.5,
self.max_w])
self.action_space = spaces.Box(low=-self.force_mag, high=self.force_mag, shape=(1,), dtype=np.float32)
self.observation_space = spaces.Box(-high, high, dtype=np.float32)
self.seed()
self.viewer = None
self.state = None
self.steps_beyond_done = None
def seed(self, seed=None):
self.np_random, seed = seeding.np_random(seed)
return [seed]
def step(self, action):
a = 0
# self.gravity = np.random.normal(10, 2)
# self.masscart = np.random.normal(1, 0.2)
# self.masspole = np.random.normal(0.1, 0.02)
self.total_mass = (self.masspole + self.masscart)
state = self.state
x, x_dot, theta, theta_dot = state
force = np.random.normal(action, 1)# wind
# force = action
costheta = math.cos(theta)
sintheta = math.sin(theta)
temp = np.random.normal((force + self.polemass_length * theta_dot * theta_dot * sintheta) / self.total_mass,0)
thetaacc = np.random.normal((self.gravity * sintheta - costheta * temp) / (
self.length * (4.0 / 3.0 - self.masspole * costheta * costheta / self.total_mass)),0)
xacc = np.random.normal(temp - self.polemass_length * thetaacc * costheta / self.total_mass,0)
if self.kinematics_integrator == 'euler':
x = x + self.tau * x_dot
x_dot = x_dot + self.tau * xacc
# x_dot = np.clip(x_dot, -self.max_v, self.max_v)
theta = theta + self.tau * theta_dot
theta_dot = theta_dot + self.tau * thetaacc
# theta_dot = np.clip(theta_dot, -self.max_w, self.max_w)
elif self.kinematics_integrator == 'friction':
xacc = -0.1 * x_dot / self.total_mass + temp - self.polemass_length * thetaacc * costheta / self.total_mass
x = x + self.tau * x_dot
x_dot = x_dot + self.tau * xacc
# x_dot = np.clip(x_dot, -self.max_v, self.max_v)
theta = theta + self.tau * theta_dot
theta_dot = theta_dot + self.tau * thetaacc
# theta_dot = np.clip(theta_dot, -self.max_w, self.max_w):
else: # semi-implicit euler
x_dot = x_dot + self.tau * xacc
x = x + self.tau * x_dot
theta_dot = theta_dot + self.tau * thetaacc
theta = theta + self.tau * theta_dot
if x>5:
x=x-10
if x<-5:
x=x+10
self.state = np.array([x, x_dot, theta, theta_dot])
done = x < -self.x_threshold \
or x > self.x_threshold \
or theta < -self.theta_threshold_radians \
or theta > self.theta_threshold_radians
done = bool(done)
if x < -self.x_threshold \
or x > self.x_threshold:
a = 1
r1 = ((self.x_threshold/5 - abs(x))) / (self.x_threshold/5) # -4-----1
r2 = ((self.theta_threshold_radians / 4) - abs(theta)) / (self.theta_threshold_radians / 4) # -3--------1
# cost1=(self.x_threshold - abs(x))/self.x_threshold
e1 = (abs(x)) / self.x_threshold
e2 = (abs(theta)) / self.theta_threshold_radians
cost = COST_V3(r1, r2, e1, e2, x, x_dot, theta, theta_dot)
return self.state, cost, done, a
def reset(self):
self.state = self.np_random.uniform(low=-0.2, high=0.2, size=(4,))
self.state[0] = self.np_random.uniform(low=-2, high=2)
# self.state[0] = 0
self.steps_beyond_done = None
return np.array(self.state)
def render(self, mode='human'):
screen_width = 800
screen_height = 400
world_width = self.x_threshold * 2
scale = screen_width / world_width
carty = 100 # TOP OF CART
polewidth = 10.0
polelen = scale * 1
cartwidth = 50.0
cartheight = 30.0
if self.viewer is None:
from gym.envs.classic_control import rendering
self.viewer = rendering.Viewer(screen_width, screen_height)
l, r, t, b = -cartwidth / 2, cartwidth / 2, cartheight / 2, -cartheight / 2
axleoffset = cartheight / 4.0
cart = rendering.FilledPolygon([(l, b), (l, t), (r, t), (r, b)])
self.carttrans = rendering.Transform()
cart.add_attr(self.carttrans)
self.viewer.add_geom(cart)
l, r, t, b = -polewidth / 2, polewidth / 2, polelen - polewidth / 2, -polewidth / 2
pole = rendering.FilledPolygon([(l, b), (l, t), (r, t), (r, b)])
pole.set_color(.8, .6, .4)
self.poletrans = rendering.Transform(translation=(0, axleoffset))
pole.add_attr(self.poletrans)
pole.add_attr(self.carttrans)
self.viewer.add_geom(pole)
self.axle = rendering.make_circle(polewidth / 2)
self.axle.add_attr(self.poletrans)
self.axle.add_attr(self.carttrans)
self.axle.set_color(.5, .5, .8)
self.viewer.add_geom(self.axle)
self.track = rendering.Line((0, carty), (screen_width, carty))
self.track.set_color(0, 0, 0)
self.viewer.add_geom(self.track)
if self.state is None: return None
x = self.state
cartx = x[0] * scale + screen_width / 2.0 # MIDDLE OF CART
self.carttrans.set_translation(cartx, carty)
self.poletrans.set_rotation(-x[2])
return self.viewer.render(return_rgb_array=mode == 'rgb_array')
def close(self):
if self.viewer:
self.viewer.close()
self.viewer = None
def COST_1000(r1, r2, e1, e2, x, x_dot, theta, theta_dot):
cost = np.sign(r2) * ((10 * r2) ** 2) - 4 * abs(x) ** 2
return cost
def COST_V3(r1, r2, e1, e2, x, x_dot, theta, theta_dot):
cost = np.sign(r2) * ((10 * r2) ** 2)
return cost
def COST_V1(r1, r2, e1, e2, x, x_dot, theta, theta_dot):
cost = np.sign(r2) * ((10 * r2) ** 2)+ np.sign(r1) * ((10 * r1) ** 2)
return cost