-
Notifications
You must be signed in to change notification settings - Fork 6
/
minecraft.py
236 lines (188 loc) · 8.86 KB
/
minecraft.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
from gym.core import Wrapper
import minerl.env.spaces as spaces
import numpy as np
from copy import deepcopy
class DummyMinecraft:
"""
Useful for debugging. (Only very limited gym compatibility is supported.)
Also used during training, since it doesn't need the actual environment.
"""
def __init__(self):
self.state = {
'equipped_items': {'mainhand': {'damage': 0, 'maxDamage': 0, 'type': 0}},
'inventory': {'coal': 0,
'cobblestone': 0,
'crafting_table': 0,
'dirt': 0,
'furnace': 0,
'iron_axe': 0,
'iron_ingot': 0,
'iron_ore': 0,
'iron_pickaxe': 0,
'log': 0,
'planks': 0,
'stick': 0,
'stone': 0,
'stone_axe': 0,
'stone_pickaxe': 0,
'torch': 0,
'wooden_axe': 0,
'wooden_pickaxe': 0},
'pov': np.full([64, 64, 3], 127, dtype=np.uint8)}
self.action_space = {
'attack': spaces.Discrete(2),
'back': spaces.Discrete(2),
'camera': spaces.Box(np.full([2], -1), np.full([2], 1)),
'craft': spaces.Enum('none', 'torch', 'stick', 'planks', 'crafting_table'),
'equip': spaces.Enum('none', 'air', 'wooden_axe', 'wooden_pickaxe', 'stone_axe',
'stone_pickaxe', 'iron_axe', 'iron_pickaxe'),
'forward': spaces.Discrete(2),
'jump': spaces.Discrete(2),
'left': spaces.Discrete(2),
'nearbyCraft': spaces.Enum('none', 'wooden_axe', 'wooden_pickaxe', 'stone_axe',
'stone_pickaxe', 'iron_axe', 'iron_pickaxe', 'furnace'),
'nearbySmelt': spaces.Enum('none', 'iron_ingot', 'coal'),
'place': spaces.Enum('none', 'dirt', 'stone', 'cobblestone', 'crafting_table', 'furnace', 'torch'),
'right': spaces.Discrete(2),
'sneak': spaces.Discrete(2),
'sprint': spaces.Discrete(2)}
self.observation_space = {
'equipped_items': {
'mainhand': {
'damage': spaces.Box(np.full([2], -1), np.full([2], 1), dtype=np.int64),
'maxDamage': spaces.Box(np.full([2], -1), np.full([2], 1), dtype=np.int64),
'type': spaces.Enum('none', 'air', 'wooden_axe', 'wooden_pickaxe', 'stone_axe', 'stone_pickaxe',
'iron_axe', 'iron_pickaxe', 'other')}},
'inventory': {
'coal': spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'cobblestone': spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'crafting_table': spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'dirt': spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'furnace': spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'iron_axe': spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'iron_ingot': spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'iron_ore': spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'iron_pickaxe': spaces.Box(np.full([1], -1), np.full([1], 1), dtype=np.int64),
'log': spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'planks': spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'stick': spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'stone': spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'stone_axe': spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'stone_pickaxe': spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'torch': spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'wooden_axe': spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64),
'wooden_pickaxe': spaces.Box(np.full([1], -1), np.full([1], -1), dtype=np.int64)},
'pov': spaces.Box(np.full([64, 64, 3], -1), np.full([64, 64, 3], -1), dtype=np.uint8)}
self.reward_range = (-np.inf, np.inf)
self.metadata = {'render.modes': ['rgb_array', 'human']}
self.t = 0
def reset(self):
self.t = 0
self.state['pov'][:, :, :] = 0
return deepcopy(self.state)
def step(self, action):
self.t += 1
self.state['pov'][:, :, :] = 0
if self.t < 1000:
return deepcopy(self.state), 0.1, False, {}
else:
return deepcopy(self.state), 0.1, True, {}
def close(self):
pass
def seed(self, _):
pass
class Env(Wrapper):
"""Main minecraft wrapper. Wraps the actions and the states with the action and state managers and creates torch
arrays on the specified device. Also activates always jumping, unless: craft, nearbyCraft, nearbySmelt,
place or attack"""
def __init__(self, env, state_manager, action_manager):
self.state_manager = state_manager
self.action_manager = action_manager
self.done = False
self.last_obs = None # used for logging
super().__init__(env)
def _process_obs(self, obs):
img, vec = self.state_manager.get_img_vec(obs)
torch_img, torch_vec = self.state_manager.get_torch_img_vec([img], [vec])
return torch_img, torch_vec
def reset(self):
obs = self.env.reset()
self.last_obs = obs
self.done = False
return self._process_obs(obs)
def step(self, action):
assert not self.done
action = self.action_manager.get_action(action)
if 'craft' in action:
if \
action['attack'] == 0 and \
action['craft'] == 0 and \
action['nearbyCraft'] == 0 and \
action['nearbySmelt'] == 0 and \
action['place'] == 0:
action['jump'] = 1
else:
if action['attack'] == 0:
action['jump'] = 1
obs, r, self.done, info = super().step(action)
self.last_obs = obs
torch_img, torch_vec = self._process_obs(obs)
return torch_img, torch_vec, r, self.done
def test_policy(writer, wrapped_env, policy, init_img, init_vec, episodes=100):
reward_list = []
steps_list = []
amount_of_episodes_with_saved_inventory = 30
first = True
i = 0
while i < episodes:
if first:
img, vec = init_img, init_vec
first = False
else:
wrapped_env.seed(i)
img, vec = wrapped_env.reset()
print("episode {}".format(i))
reward = 0.
frame_steps = 0
steps = 0
last_obs = wrapped_env.last_obs
last_meta_id = 0
if 'inventory' in last_obs:
last_print_dict = deepcopy(last_obs['inventory'])
last_print_dict['meta_id'] = last_meta_id
if i < amount_of_episodes_with_saved_inventory:
writer.add_scalars(f"episode {i} inventory", last_print_dict, steps)
done = False
while not done:
a_id = policy(img, vec)
img, vec, r, done = wrapped_env.step(a_id)
reward += r
steps += 1
frame_steps += 1
obs = wrapped_env.last_obs
if 'inventory' in last_obs:
print_dict = deepcopy(obs['inventory'])
if i < amount_of_episodes_with_saved_inventory:
if print_dict != last_print_dict:
writer.add_scalars(f"episode {i} inventory", print_dict, steps)
last_print_dict = print_dict
tmp_p = 6000
if steps % tmp_p == 0:
print(f"{frame_steps} / {18000}")
if steps == 1:
print("always terminal bug detected")
raise RuntimeError
else:
reward_list.append(reward)
steps_list.append(steps)
print(f"episode reward: {reward} , episode terminated after {steps} env steps")
print(f"avg_reward after {i + 1} (out of {episodes}) episodes: {np.mean(reward_list)}")
writer.add_scalar("reward", reward, i)
writer.add_scalar("steps", steps, i)
writer.flush()
i += 1
print("Total avg_reward: {}".format(np.mean(reward_list)))
writer.add_scalar("avg_reward", np.mean(reward_list), 0)
writer.add_scalar("avg_steps", np.mean(steps_list), 0)
writer.flush()
writer.close()