forked from openai/spinningup
-
Notifications
You must be signed in to change notification settings - Fork 0
/
rrc_utils.py
315 lines (252 loc) · 14.4 KB
/
rrc_utils.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
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
import gym
import numpy as np
from gym import wrappers
import functools
from gym.envs.registration import register
phase = 2
if phase == 1:
from rrc_simulation.gym_wrapper.envs import cube_env, custom_env
elif phase == 2:
from rrc_iprl_package.envs import cube_env
from rrc_iprl_package.envs import custom_env
from rrc_iprl_package.envs import env_wrappers
registered_envs = [spec.id for spec in gym.envs.registry.all()]
FRAMESKIP = 10
EPLEN = 120 * 1000 // FRAMESKIP # 15 seconds
EPLEN_SHORT = 5 * 1000 // FRAMESKIP # 5 seconds, 500 total timesteps
if phase == 1:
if "real_robot_challenge_phase_1-v2" not in registered_envs:
register(
id="real_robot_challenge_phase_1-v2",
entry_point=custom_env.PushCubeEnv
)
if "real_robot_challenge_phase_1-v3" not in registered_envs:
register(
id="real_robot_challenge_phase_1-v3",
entry_point=custom_env.PushReorientCubeEnv
)
if "real_robot_challenge_phase_1-v4" not in registered_envs:
register(
id="real_robot_challenge_phase_1-v4",
entry_point=custom_env.SparseCubeEnv
)
elif phase == 2:
if "real_robot_challenge_phase_2-v1" not in registered_envs:
register(
id="real_robot_challenge_phase_2-v1",
entry_point=cube_env.RealRobotCubeEnv
)
if "real_robot_challenge_phase_2-v2" not in registered_envs:
register(
id="real_robot_challenge_phase_2-v2",
entry_point=cube_env.PushCubeEnv
)
total_steps = 5e6
step_rates = np.linspace(0, 0.6, 10)
def success_rate_early_stopping(steps, success_rate):
return step_rates[min(9, int(steps/total_steps * 10))] > success_rate
def make_env_fn(env_str, wrapper_params=[], **make_kwargs):
"""Returns env_fn to pass to spinningup alg"""
def env_fn(visualization=False):
env = gym.make(env_str, visualization=visualization, **make_kwargs)
for w in wrapper_params:
if isinstance(w, dict):
env = w['cls'](env, *w.get('args', []), **w.get('kwargs', {}))
else:
env = w(env)
return env
return env_fn
if phase == 1:
push_random_initializer = cube_env.RandomInitializer(difficulty=1)
fixed_reorient_initializer = custom_env.RandomGoalOrientationInitializer(difficulty=1)
push_curr_initializer = custom_env.CurriculumInitializer(initial_dist=0.,
num_levels=5)
push_fixed_initializer = custom_env.CurriculumInitializer(initial_dist=0.,
num_levels=2)
reorient_initializer = reorient_curr_initializer = custom_env.CurriculumInitializer(
initial_dist=0.06, num_levels=3, difficulty=4,
fixed_goal=custom_env.RandomOrientationInitializer.goal)
recenter_initializer = custom_env.ReorientInitializer(1, 0.09)
push_initializer = push_fixed_initializer
lift_initializer = cube_env.RandomInitializer(difficulty=2)
ori_initializer = cube_env.RandomInitializer(difficulty=3)
# Val in info string calls logger.log_tabular() with_min_and_max to False
push_info_kwargs = {'is_success': 'SuccessRateVal', 'final_dist': 'FinalDist',
'final_score': 'FinalScore', 'init_sample_radius': 'InitSampleDistVal',
'goal_sample_radius': 'GoalSampleDistVal'}
reorient_info_kwargs = {'is_success': 'SuccessRateVal',
'is_success_ori': 'OriSuccessRateVal',
'final_dist': 'FinalDist', 'final_ori_dist': 'FinalOriDist',
'final_ori_scaled': 'FinalOriScaledDist',
'final_score': 'FinalScore'}
info_keys = ['is_success', 'is_success_ori', 'final_ori_dist', 'final_dist',
'final_score']
curr_info_keys = info_keys + ['goal_sample_radius', 'init_sample_radius']
reorient_info_keys = ['is_success', 'is_success_ori', 'final_dist', 'final_score',
'final_ori_dist', 'final_ori_scaled']
action_type = cube_env.ActionType.POSITION
log_info_wrapper = functools.partial(custom_env.LogInfoWrapper,
info_keys=info_keys)
reorient_log_info_wrapper = functools.partial(custom_env.LogInfoWrapper,
info_keys=reorient_info_keys)
final_wrappers = [functools.partial(wrappers.TimeLimit, max_episode_steps=EPLEN),
log_info_wrapper,
wrappers.ClipAction, wrappers.FlattenObservation]
final_wrappers = final_wrappers_short = [
functools.partial(wrappers.TimeLimit, max_episode_steps=EPLEN_SHORT),
log_info_wrapper,
wrappers.FlattenObservation]
final_wrappers_reorient = [
functools.partial(wrappers.TimeLimit, max_episode_steps=EPLEN_SHORT),
reorient_log_info_wrapper,
wrappers.FlattenObservation]
final_wrappers_reorient_abs = [
functools.partial(wrappers.TimeLimit, max_episode_steps=EPLEN_SHORT),
reorient_log_info_wrapper,
wrappers.FlattenObservation]
final_wrappers_relgoal = [
functools.partial(custom_env.RelativeGoalWrapper,
keep_goal=False),
functools.partial(wrappers.TimeLimit, max_episode_steps=EPLEN_SHORT),
reorient_log_info_wrapper,
wrappers.FlattenObservation]
final_wrappers_vds = [functools.partial(wrappers.TimeLimit, max_episode_steps=EPLEN),
custom_env.FlattenGoalWrapper]
abs_scaled_wrapper = functools.partial(custom_env.ScaledActionWrapper,
goal_env=False, relative=False)
rel_scaled_wrapper = functools.partial(custom_env.ScaledActionWrapper,
goal_env=False, relative=True)
abs_task_wrapper = functools.partial(custom_env.TaskSpaceWrapper, relative=False)
rel_task_wrapper = functools.partial(custom_env.TaskSpaceWrapper, relative=True)
rew_wrappers_step = [functools.partial(custom_env.CubeRewardWrapper,
pos_coef=1., ori_coef=1.,
ac_norm_pen=0.2, fingertip_coef=1.,
rew_fn='exp', augment_reward=True),
custom_env.StepRewardWrapper,
functools.partial(custom_env.ReorientWrapper,
goal_env=False, dist_thresh=0.075,
ori_thresh=np.pi/6)]
rew_wrappers = [functools.partial(custom_env.CubeRewardWrapper,
pos_coef=.1, ori_coef=.1,
ac_norm_pen=.1, fingertip_coef=.1,
rew_fn='exp', augment_reward=True),
functools.partial(custom_env.ReorientWrapper,
goal_env=False, dist_thresh=0.075,
ori_thresh=np.pi)]
recenter_rew_wrappers = [functools.partial(custom_env.CubeRewardWrapper, pos_coef=1.,
ori_coef=.5, ac_norm_pen=0., augment_reward=True, rew_fn='exp'),
functools.partial(custom_env.ReorientWrapper, goal_env=False,
dist_thresh=0.05,
ori_thresh=np.pi),
rel_scaled_wrapper]
rrc_wrappers = [rel_scaled_wrapper] + final_wrappers
rrc_vds_wrappers = [rel_scaled_wrapper] + final_wrappers_vds
push_wrappers = [functools.partial(custom_env.CubeRewardWrapper, pos_coef=1.,
ac_norm_pen=0.2, rew_fn='exp'),
rel_scaled_wrapper]
push_wrappers = push_wrappers + final_wrappers
recenter_wrappers_rel = recenter_rew_wrappers + final_wrappers_reorient
reorient_wrappers_relgoal = recenter_rew_wrappers + final_wrappers_relgoal
reorient_wrappers_relgoaltask = [rel_task_wrapper] + rew_wrappers + final_wrappers_relgoal
recenter_wrappers_abs = recenter_rew_wrappers + final_wrappers_reorient
abs_task_wrappers = [abs_task_wrapper] + rew_wrappers + final_wrappers_reorient + [wrappers.ClipAction]
rel_task_wrappers = [rel_task_wrapper] + rew_wrappers + final_wrappers_reorient + [wrappers.ClipAction]
abs_task_step_wrappers = [abs_task_wrapper] + rew_wrappers_step + final_wrappers_reorient + [wrappers.ClipAction]
rel_task_step_wrappers = [rel_task_wrapper] + rew_wrappers_step + final_wrappers_reorient + [wrappers.ClipAction]
rrc_env_str = 'rrc_simulation.gym_wrapper:real_robot_challenge_phase_1-v1'
rrc_env_fn = make_env_fn(rrc_env_str, rrc_wrappers,
initializer=push_initializer,
action_type=action_type,
frameskip=FRAMESKIP)
rrc_vds_env_fn = make_env_fn(rrc_env_str, rrc_vds_wrappers,
initializer=push_initializer,
action_type=action_type,
frameskip=FRAMESKIP)
push_env_str = 'real_robot_challenge_phase_1-v2'
push_env_fn = make_env_fn(push_env_str, push_wrappers,
initializer=push_initializer,
action_type=action_type,
frameskip=FRAMESKIP)
reorient_env_str = 'real_robot_challenge_phase_1-v3'
abs_task_env_fn = make_env_fn(reorient_env_str, abs_task_wrappers,
initializer=reorient_initializer,
action_type=cube_env.ActionType.TORQUE,
frameskip=5)
rel_task_env_fn = make_env_fn(reorient_env_str, rel_task_wrappers,
initializer=reorient_initializer,
action_type=cube_env.ActionType.TORQUE,
frameskip=5)
abs_task_step_env_fn = make_env_fn(reorient_env_str, abs_task_step_wrappers,
initializer=reorient_initializer,
action_type=cube_env.ActionType.TORQUE,
frameskip=5)
rel_task_step_env_fn = make_env_fn(reorient_env_str, rel_task_step_wrappers,
initializer=reorient_initializer,
action_type=cube_env.ActionType.TORQUE,
frameskip=5)
recenter_rel_env_fn = make_env_fn(reorient_env_str, recenter_wrappers_rel,
initializer=recenter_initializer,
action_type=action_type,
frameskip=FRAMESKIP)
recenter_env_fn = make_env_fn(reorient_env_str, recenter_wrappers_abs,
initializer=recenter_initializer,
action_type=action_type,
frameskip=FRAMESKIP)
reorient_env_fn = make_env_fn(reorient_env_str, reorient_wrappers_relgoal,
initializer=fixed_reorient_initializer,
action_type=action_type,
frameskip=FRAMESKIP)
reorient_task_env_fn = make_env_fn(reorient_env_str, reorient_wrappers_relgoaltask,
initializer=fixed_reorient_initializer,
action_type=cube_env.ActionType.TORQUE,
frameskip=5)
eval_keys = ['is_success', 'is_success_ori', 'final_ori_dist', 'final_dist',
'final_score']
# PHASE 2
p2_fixed_reorient = env_wrappers.RandomGoalOrientationInitializer(difficulty=1)
p2_push_curr = env_wrappers.CurriculumInitializer(initial_dist=0.,
num_levels=5)
p2_push_fixed = env_wrappers.CurriculumInitializer(initial_dist=0.,
num_levels=2)
p2_reorient_curr = env_wrappers.CurriculumInitializer(
initial_dist=0.06, num_levels=3, difficulty=4,
fixed_goal=env_wrappers.RandomOrientationInitializer.goal)
p2_recenter = env_wrappers.ReorientInitializer(1, 0.09)
p2_push = p2_push_fixed
p2_goalenv_str = "real_robot_challenge_phase_2-v2"
p2_env_str = "real_robot_challenge_phase_2-v2"
p2_info_keys = ['is_success', 'is_success_ori', 'final_dist', 'final_score',
'final_ori_dist', 'final_ori_scaled']
p2_log_info_wrapper = functools.partial(env_wrappers.LogInfoWrapper,
info_keys=p2_info_keys)
p2_final_wrappers = [functools.partial(wrappers.TimeLimit, max_episode_steps=EPLEN),
p2_log_info_wrapper,
wrappers.ClipAction, wrappers.FlattenObservation]
p2_final_wrappers_reorient = [
functools.partial(wrappers.TimeLimit, max_episode_steps=EPLEN_SHORT),
p2_log_info_wrapper,
wrappers.FlattenObservation]
p2_final_wrappers_relgoal = [functools.partial(env_wrappers.RelativeGoalWrapper,
keep_goal=False)] + p2_final_wrappers_reorient
p2_rew_wrappers = [functools.partial(env_wrappers.CubeRewardWrapper,
pos_coef=.1, ori_coef=.1,
ac_norm_pen=.1, fingertip_coef=.1,
rew_fn='exp', augment_reward=True),
functools.partial(env_wrappers.ReorientWrapper,
goal_env=False, dist_thresh=0.075,
ori_thresh=np.pi)]
p2_rel_scaled_wrapper = functools.partial(env_wrappers.ScaledActionWrapper,
goal_env=False, relative=True)
p2_recenter_rew_wrappers = [functools.partial(env_wrappers.CubeRewardWrapper, pos_coef=1.,
ori_coef=.5, ac_norm_pen=0., augment_reward=True, rew_fn='exp'),
functools.partial(env_wrappers.ReorientWrapper, goal_env=False,
dist_thresh=0.05,
ori_thresh=np.pi),
p2_rel_scaled_wrapper]
p2_rrc_wrappers = [p2_rel_scaled_wrapper] + p2_final_wrappers_relgoal
p2_reorient_env_fn = make_env_fn(
p2_env_str,
[p2_rel_scaled_wrapper] + p2_final_wrappers_relgoal,
initializer=p2_fixed_reorient,
action_type=cube_env.ActionType.TORQUE,
frameskip=5)