-
Notifications
You must be signed in to change notification settings - Fork 0
/
custom_snake_env.py
279 lines (228 loc) · 9.75 KB
/
custom_snake_env.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
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
import math
from copy import deepcopy
from random import randint
import gym as gym
import numpy as np
import pygame as pygame
from gym import spaces
def move_to_direction(pos, direction):
x, y = pos
x = x + ((1 - direction) if (direction % 2 == 0) else 0)
y = y + ((direction - 2) if (direction % 2 != 0) else 0)
return x, y
def generate_unique(generated, range):
x = randint(0, range[0] - 1)
y = randint(0, range[1] - 1)
while (x, y) in generated:
x = randint(0, range[0] - 1)
y = randint(0, range[1] - 1)
return x, y
class SnakeGame(gym.Env):
AGENT_COLORS = [
(0, 255, 0),
(0, 0, 255),
(255, 255, 0),
(255, 0, 255),
]
class Agent:
def __init__(self):
self.direction = randint(0, 3)
self.head = (0, 0)
self.body = [(0, 0)]
self.dead = False
def reset(self, x, y):
self.direction = randint(0, 3)
self.head = (x, y)
self.body = [(x, y)]
self.dead = False
def move(self, action, ate_food):
self.direction = (self.direction + action) % 4
moved = move_to_direction(self.head, self.direction)
self.head = moved
self.body.insert(0, deepcopy(moved))
if not ate_food:
self.body.pop()
def draw(self, scale, surface, image):
for pos in self.body:
surface.blit(image, ((pos[0]+1)*scale, (pos[1]+1)*scale))
def __init__(self, num_agents=2, num_fruits=20, width=30, height=30):
self.agents = [self.Agent() for _ in range(num_agents)]
self.fruits = []
self.num_agents = num_agents
self.num_fruits = num_fruits
self.width = width
self.height = height
self.scale = 15
self.display_width = (width+2) * self.scale
self.display_height = (height+2) * self.scale
self.steps = 0
self.max_steps = 1000
self._display_surf = None
self._image_surf = None
self._fruit_surf = None
self.reset()
self._pygame_init()
self.observation_space = spaces.Box(low=0, high=4, shape=(self.num_agents, 2, 3))
self.action_space = spaces.Tuple(
[spaces.Discrete(3) for i in range(self.num_agents)]
)
def step(self, actions):
rewards = [-0.1] * self.num_agents
actions = [a-1 for a in actions]
for i, agent in enumerate(self.agents):
if agent.dead:
continue
preview = move_to_direction(agent.head, (agent.direction+actions[i])%4)
head_snapshot = deepcopy(agent.head)
ate_food = preview in self.fruits
agent.move(actions[i], ate_food)
closest = self.fruits[np.argmin([math.dist(f, agent.head) for f in self.fruits])]
if math.dist(head_snapshot, closest) > math.dist(agent.head, closest):
rewards[i] = 0.1
self.fruits = [x for x in self.fruits if x != preview]
if ate_food:
rewards[i] = 10
if len(self.fruits) < self.num_fruits:
# self.fruits = []
generated = list(self.fruits)
num_more = self.num_fruits - len(self.fruits)
for i in range(num_more):
pair = generate_unique(generated, (self.width, self.height))
self.fruits.append(pair)
generated.append(pair)
continue
x, y = agent.head
if x < 0 or x >= self.width or y < 0 or y >= self.height:
rewards[i] = -10
agent.dead = True
continue
if agent.head in agent.body[1:]:
rewards[i] = -10
agent.dead = True
continue
for i, agent in enumerate(self.agents):
if agent.dead:
continue
collides = False
for j, snake in enumerate(self.agents):
if i == j:
continue
if agent.head in snake.body:
collides = True
break
agent.dead = collides
if collides:
rewards[i] = -10
continue
done = sum([1 for x in self.agents if not x.dead]) == 0
done = done or len(self.fruits) == 0
self.steps = self.steps + 1
if self.steps > self.max_steps:
done = True
return self.get_new_relative_state(), rewards, done, {}
def reset(self):
self.steps = 0
generated = []
for agent in self.agents:
x, y = generate_unique(generated, (self.width, self.height))
agent.reset(x, y)
generated.append((x, y))
self.fruits = []
for i in range(self.num_fruits):
pair = generate_unique(generated, (self.width, self.height))
self.fruits.append(pair)
generated.append(pair)
return self.get_new_relative_state()
# Test
def get_new_relative_state(self):
states = []
for agent in self.agents:
other_snake_bodies = [b for b in [a.body for a in self.agents if a != agent]][0]
closest = self.fruits[np.argmin([math.dist(f, agent.head) for f in self.fruits])]
x, y = agent.head
bodies_right = sum([1 for b in other_snake_bodies if 0 < b[0] - x < 3 and math.dist(b, agent.head) < 5]) > 1
bodies_left = sum([1 for b in other_snake_bodies if 0 < x - b[0] < 3 and math.dist(b, agent.head) < 5]) > 1
bodies_down = sum([1 for b in other_snake_bodies if 0 < b[1] - y < 3 and math.dist(b, agent.head) < 5]) > 1
bodies_up = sum([1 for b in other_snake_bodies if 0 < y - b[1] < 3 and math.dist(b, agent.head) < 5]) > 1
state = (
int(closest[0] > x),
int(closest[0] < x),
int(closest[1] > y),
int(closest[1] < y),
int(x == 0),
int(y == 0),
int(x == self.width - 1),
int(y == self.height - 1),
int(bodies_left or (x - 1, y) in agent.body),
int(bodies_up or (x, y - 1) in agent.body),
int(bodies_right or (x + 1, y) in agent.body),
int(bodies_down or (x, y + 1) in agent.body),
int(agent.direction == 0),
int(agent.direction == 1),
int(agent.direction == 2),
int(agent.direction == 3),
)
states.append(state)
return np.array(states)
def get_povs(self):
povs = []
for a, agent in enumerate(self.agents):
vx = move_to_direction((0, 0), agent.direction)
vy = move_to_direction((0, 0), (agent.direction-1) % 4)
pov = np.zeros((5, 7))
hx, hy = agent.head
for i in range(5):
for j in range(7):
x = hx + vx[0]*i + vy[0]*(3-j)
y = hy + vx[1]*i + vy[1]*(3-j)
if x < 0 or x >= self.width or y < 0 or y >= self.height:
pov[i][j] = -1
elif (x, y) in self.fruits:
pov[i][j] = 1
elif (x, y) in agent.body:
pov[i][j] = -1
else:
other = False
for other_a, other_agent in enumerate(self.agents):
if a == other_a:
continue
if (x, y) in other_agent.body:
other = True
break
pov[i][j] = -1 if other else 0
povs.append(pov)
return povs
def close(self):
pygame.quit()
def render(self):
self._pygame_init()
self._draw_env()
for f in self.fruits:
self._pygame_draw(self._display_surf, self._fruit_surf, f)
for i, a in enumerate(self.agents):
if a.dead:
continue
a.draw(self.scale, self._display_surf, self._agent_surfs[i])
pygame.display.flip()
def _draw_env(self):
self._display_surf.fill((0, 110, 110))
for i in range(0, self.display_width, self.scale):
self._display_surf.blit(self._wall_surf, (0, i))
self._display_surf.blit(self._wall_surf, (self.display_width - self.scale, i))
for i in range(0, self.display_width, self.scale):
self._display_surf.blit(self._wall_surf, (i, 0))
self._display_surf.blit(self._wall_surf, (i, self.display_width - self.scale))
def _pygame_draw(self, surface, image, pos):
surface.blit(image, ((pos[0]+1)*self.scale, (pos[1]+1)*self.scale))
def _pygame_init(self):
pygame.init()
self._display_surf = pygame.display.set_mode((self.display_width, self.display_height), pygame.HWSURFACE)
self._agent_surfs = []
for i, p in enumerate(self.agents):
image_surf = pygame.Surface([self.scale, self.scale])
image_surf.fill(self.AGENT_COLORS[i % len(self.AGENT_COLORS)])
self._agent_surfs.append(image_surf)
self._fruit_surf = pygame.Surface([self.scale, self.scale])
self._fruit_surf.fill((255, 0, 0))
self._wall_surf = pygame.Surface([self.scale, self.scale])
self._wall_surf.fill((255, 255, 255))