-
Notifications
You must be signed in to change notification settings - Fork 0
/
gridworld.py
178 lines (138 loc) · 4.46 KB
/
gridworld.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
#!/usr/bin/env python3
from enum import Enum
from mdp import FiniteStateMDP, MDPState
import itertools
import numpy as np
class Actions(Enum):
UP=1
DOWN=2
LEFT=3
RIGHT=4
_TXT = {
Actions.UP: "^^",
Actions.DOWN: "vv",
Actions.LEFT: "<<",
Actions.RIGHT: ">>",
}
_UP = np.array([0, 1])
_DOWN = np.array([0, -1])
_LEFT = np.array([-1, 0])
_RIGHT = np.array([1, 0])
class GridState(MDPState):
def __init__(self, x, y, width, height):
self.x = x
self.y = y
self.has_gold = has_gold
self.has_immunity = has_immunity
self._width = width
self._height = height
def clone(self):
return GridState(self.x, self.y, self._width, self._height)
@property
def pos(self):
return np.array([self.x, self.y])
@property
def i(self):
return self._height * self.x + self.y
def __repr__(self):
return '({x}, {y})'.format(**self.__dict__)
def _clip(p, max_x, max_y):
p = np.array([max(min(p[0], max_x), 1),
max(min(p[1], max_y), 1)])
return p
_OBS_KEYS = ['pit']
_OBS_REWARDS = {
'pit': -10.0,
'goal': 10.0
}
_OBJ_KEYS = []
class DiscreteGridWorldMDP(FiniteStateMDP):
def __init__(self, w, h, move_cost=-0.1):
self._w = w
self._h = h
self.move_cost = move_cost
self._obs = {k:{} for k in _OBS_KEYS}
@property
def width(self):
return self._w
@property
def height(self):
return self._h
@property
def num_states(self):
return self.width * self.height * 4
@property
def states(self):
return itertools.product(
range(self.width), range(self.height), (True, False), (True, False))
@property
def actions(self):
return Actions
@property
def initial_state(self):
return WumpusState(0, 0, False, False, self.width, self.height)
def actions_at(self, state):
a = [Actions.LEFT, Actions.RIGHT, Actions.UP, Actions.DOWN]
return a
def p(self, state, action):
if action in [Actions.UP, Actions.DOWN, Actions.LEFT, Actions.RIGHT]:
return self.move(state, action)
else:
raise Exception("Invalid action specified: {}".format(action))
def r(self, s1, s2):
## if it's the pit, return the default pit bad reward
if self.obs_at('pit', s2.pos):
return self._obs['pit'][tuple(s2.pos)]
## if it's the goal, return the goal reward + gold reward if applicable
if self.obs_at('goal', s2.pos):
return self._obs['goal'][tuple(s2.pos)]
## otherwise return the move cost
return self.move_cost
def is_terminal(self, state):
## if we're at the wumpus and have no immunity, we die
if self.obs_at('pit', state.pos):
return True
## if we're at the goal, we win
if self.obs_at('goal', state.pos):
return True
return False
def obs_at(self, kind, pos):
return tuple(pos) in self._obs[kind].keys()
def move(self, state, action):
probs = [0.7, 0.15, 0.15]
if action == Actions.UP:
alst = [_UP, _LEFT, _RIGHT]
elif action == Actions.DOWN:
alst = [_DOWN, _RIGHT, _LEFT]
elif action == Actions.LEFT:
alst = [_LEFT, _UP, _DOWN]
elif action == Actions.RIGHT:
alst = [_RIGHT, _DOWN, _UP]
x = []
for a in alst:
new_state = state.clone()
new_pos = _clip(state.pos + a, self.width, self.height)
new_state.x = new_pos[0]
new_state.y = new_pos[1]
x += [new_state]
return zip(x, probs)
def add_obstacle(self, kind, pos, reward=None):
## default rewards
if not reward:
reward = _OBS_REWARDS[kind]
self._obs[kind][tuple(pos)] = reward
def display(self):
obs_lab = lambda p, lab, kind: lab if self.obs_at(kind, p) else ' '
print(' ', end='')
for i in range(self.width):
print(' {:5d} '.format(i),end='')
print()
for j in reversed(range(self.height)):
print('{:5d} '.format(j), end='')
for i in range(self.width):
p = tuple([i, j])
l_s = 'S' if p == (0, 0) else ' '
l_p = obs_lab(p, 'P', 'pit')
l_gl = obs_lab(p, 'X', 'goal')
print('|' + l_s+l_w+l_p+l_gl+l_gd+l_i,end='')
print('|')