-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathcrafter_env.py
204 lines (175 loc) · 8.18 KB
/
crafter_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
""" Create crafter environment """
import copy
from collections import deque
import dm_env
import gym
import cv2
import numpy as np
from dm_env import specs, StepType
import text_crafter.text_crafter
from utils import ExtendedTimeStepWrapper
from text_crafter.text_crafter.logging_wrapper import CrafterLoggingWrapper
from text_crafter.text_crafter.goal_wrapper import CrafterGoalWrapper, CrafterLMGoalWrapper
class Crafter(dm_env.Environment):
"""A Crafter env, wrapped to do logging."""
def __init__(self,
logdir,
env_spec,
screen_size=84,
save_stats=True,
save_video=False,
save_episode=False,
seed=1,
env_reward=False,
use_wandb=False,
debug=False,
device=None):
if env_spec['name'] == 'CrafterReward-v1':
assert env_reward
env_spec['env_reward'] = env_reward
env_spec['device'] = device
env_spec['seed'] = seed
self.logdir = logdir
env = gym.make(env_spec['name'], **env_spec)
if 'CrafterTextEnv' in env_spec['name']: # NOT baseline, so we have goals and should log
env = CrafterLoggingWrapper(CrafterLMGoalWrapper(env,
env_spec['lm_spec'],
env_spec['env_reward'],
device=device,
threshold=env_spec['threshold'],
debug=debug,
single_task=env_spec['single_task'],
single_goal_hierarchical=env_spec['single_goal_hierarchical'],
use_state_captioner=env_spec['use_state_captioner'],
use_transition_captioner=env_spec['use_transition_captioner'],
check_ac_success=env_spec['check_ac_success']
))
else:
env = CrafterGoalWrapper(env, env_spec['env_reward'], env_spec['single_task'], single_goal_hierarchical=env_spec['single_goal_hierarchical'])
self._env = text_crafter.text_crafter.Recorder(
env,
logdir,
save_stats=save_stats,
save_video=save_video,
save_episode=save_episode,
use_wandb=use_wandb)
self._env.seed(seed)
self._screen_size = screen_size
shape = (1, screen_size, screen_size)
self._obs_spec = {'obs':specs.BoundedArray(shape=shape,
dtype=np.uint8,
minimum=0,
maximum=255,
name='observation'),
'text_obs':specs.Array(shape=(env_spec['max_seq_len'],),
dtype=np.uint8,
name='text_obs'),
'goal':specs.Array(shape=(env_spec['max_seq_len'],),
dtype=np.uint8,
name='goal'),
'old_goals':specs.Array(shape=(env_spec['max_seq_len'],),
dtype=np.uint8,
name='old_goals'),
'success':specs.Array(shape=(),
dtype=bool,
name='success'),
'goal_success':specs.Array(shape=(),
dtype=bool,
name='goal_success'),
}
self._action_spec = specs.DiscreteArray(
num_values=self._env.action_space.n,
dtype=np.int64,
name='action')
self._env_reward = None
def _transform_observation(self, obs):
obs['obs'] = self._transform_obs_array(obs['obs'])
return obs
def _transform_obs_array(self, obs):
# gray scale
obs = np.mean(obs, axis=-1)
# resize
image = cv2.resize(obs, (self._screen_size, self._screen_size),
interpolation=cv2.INTER_LINEAR)
obs = np.asarray(image, dtype=np.uint8)
obs = np.expand_dims(obs, axis=0)
return obs
def get_env_reward(self):
"""Return the most recent env reward."""
return self._env_reward
def reset(self):
obs, info = self._env.reset()
self._env_reward = None
obs = self._transform_observation(obs)
return dm_env.TimeStep(StepType.FIRST, 0.0, 1.0, obs), info # StepType, reward, discount, obs
def step(self, action):
obs, reward, done, info = self._env.step(action)
obs = self._transform_observation(obs)
self._env_reward = info['env_reward']
if done:
return dm_env.termination(reward, obs)
return dm_env.transition(reward, obs)
def observation_spec(self):
return self._obs_spec
def action_spec(self):
return self._action_spec
def reward_spec(self):
return specs.Array(shape=(), name='reward', dtype=np.float16)
def discount_spec(self):
return specs.Array(shape=(), name='discount', dtype=np.uint8)
def __getattr__(self, name):
if name.startswith('__'):
raise AttributeError(name)
return getattr(self._env, name)
class FrameStack(dm_env.Environment):
"""A dm_env wrapper that stacks a list of the past k observations."""
def __init__(self, env, k):
self._env = env
self._k = k
self._frames = deque([], maxlen=k)
env_obs_spec = env.observation_spec()
obs_shape = env_obs_spec['obs'].shape
env_obs_spec['obs'] = specs.BoundedArray(shape=np.concatenate(
[[obs_shape[0] * k], obs_shape[1:]], axis=0),
dtype=np.uint8,
minimum=0,
maximum=255,
name='observation')
def _transform_observation(self, time_step):
assert len(self._frames) == self._k
frame_list = list(self._frames)
if isinstance(frame_list[0], dict):
obs = copy.deepcopy(frame_list[-1])
obs['obs'] = np.concatenate([o['obs'] for o in frame_list], axis=0)
else:
obs = np.concatenate(frame_list, axis=0)
return time_step._replace(observation=obs)
def reset(self):
time_step, info = self._env.reset()
pixels = time_step.observation
for _ in range(self._k):
self._frames.append(pixels)
return self._transform_observation(time_step), info
def step(self, action):
time_step = self._env.step(action)
pixels = time_step.observation
self._frames.append(pixels)
return self._transform_observation(time_step)
def observation_spec(self):
return self._obs_spec
def action_spec(self):
return self._env.action_spec()
def reward_spec(self):
return self._env.reward_spec()
def discount_spec(self):
return self._env.discount_spec()
def __getattr__(self, name):
return getattr(self._env, name)
def make(logdir, env_spec, save_video, frame_stack, seed=1, env_reward=False,
use_wandb=False, debug=False, device=None):
"""Create and wrap crafter environment."""
env = Crafter(logdir, env_spec, save_video=save_video, seed=seed, env_reward=env_reward,
use_wandb=use_wandb, debug=debug, device=device)
env = FrameStack(env, k=frame_stack)
env = ExtendedTimeStepWrapper(env)
return env