forked from google-deepmind/dm_control
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathhumanoid.py
More file actions
207 lines (171 loc) · 7.77 KB
/
humanoid.py
File metadata and controls
207 lines (171 loc) · 7.77 KB
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
# Copyright 2017 The dm_control Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
"""Humanoid Domain."""
import collections
from dm_control import mujoco
from dm_control.rl import control
from dm_control.suite import base
from dm_control.suite import common
from dm_control.suite.utils import randomizers
from dm_control.utils import containers
from dm_control.utils import rewards
import numpy as np
_DEFAULT_TIME_LIMIT = 25
_CONTROL_TIMESTEP = .025
# Height of head above which stand reward is 1.
_STAND_HEIGHT = 1.4
# Horizontal speeds above which move reward is 1.
_WALK_SPEED = 1
_RUN_SPEED = 10
SUITE = containers.TaggedTasks()
def get_model_and_assets():
"""Returns a tuple containing the model XML string and a dict of assets."""
return common.read_model('humanoid.xml'), common.ASSETS
@SUITE.add('benchmarking')
def stand(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns the Stand task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Humanoid(move_speed=0, pure_state=False, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, control_timestep=_CONTROL_TIMESTEP,
**environment_kwargs)
@SUITE.add('benchmarking')
def walk(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns the Walk task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Humanoid(move_speed=_WALK_SPEED, pure_state=False, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, control_timestep=_CONTROL_TIMESTEP,
**environment_kwargs)
@SUITE.add('benchmarking')
def run(time_limit=_DEFAULT_TIME_LIMIT, random=None, environment_kwargs=None):
"""Returns the Run task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Humanoid(move_speed=_RUN_SPEED, pure_state=False, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, control_timestep=_CONTROL_TIMESTEP,
**environment_kwargs)
@SUITE.add()
def run_pure_state(time_limit=_DEFAULT_TIME_LIMIT, random=None,
environment_kwargs=None):
"""Returns the Run task."""
physics = Physics.from_xml_string(*get_model_and_assets())
task = Humanoid(move_speed=_RUN_SPEED, pure_state=True, random=random)
environment_kwargs = environment_kwargs or {}
return control.Environment(
physics, task, time_limit=time_limit, control_timestep=_CONTROL_TIMESTEP,
**environment_kwargs)
class Physics(mujoco.Physics):
"""Physics simulation with additional features for the Walker domain."""
def torso_upright(self):
"""Returns projection from z-axes of torso to the z-axes of world."""
return self.named.data.xmat['torso', 'zz']
def head_height(self):
"""Returns the height of the torso."""
return self.named.data.xpos['head', 'z']
def center_of_mass_position(self):
"""Returns position of the center-of-mass."""
return self.named.data.subtree_com['torso'].copy()
def center_of_mass_velocity(self):
"""Returns the velocity of the center-of-mass."""
return self.named.data.sensordata['torso_subtreelinvel'].copy()
def torso_vertical_orientation(self):
"""Returns the z-projection of the torso orientation matrix."""
return self.named.data.xmat['torso', ['zx', 'zy', 'zz']]
def joint_angles(self):
"""Returns the state without global orientation or position."""
return self.data.qpos[7:].copy() # Skip the 7 DoFs of the free root joint.
def extremities(self):
"""Returns end effector positions in egocentric frame."""
torso_frame = self.named.data.xmat['torso'].reshape(3, 3)
torso_pos = self.named.data.xpos['torso']
positions = []
for side in ('left_', 'right_'):
for limb in ('hand', 'foot'):
torso_to_limb = self.named.data.xpos[side + limb] - torso_pos
positions.append(torso_to_limb.dot(torso_frame))
return np.hstack(positions)
class Humanoid(base.Task):
"""A humanoid task."""
def __init__(self, move_speed, pure_state, random=None):
"""Initializes an instance of `Humanoid`.
Args:
move_speed: A float. If this value is zero, reward is given simply for
standing up. Otherwise this specifies a target horizontal velocity for
the walking task.
pure_state: A bool. Whether the observations consist of the pure MuJoCo
state or includes some useful features thereof.
random: Optional, either a `numpy.random.RandomState` instance, an
integer seed for creating a new `RandomState`, or None to select a seed
automatically (default).
"""
self._move_speed = move_speed
self._pure_state = pure_state
super().__init__(random=random)
def initialize_episode(self, physics):
"""Sets the state of the environment at the start of each episode.
Args:
physics: An instance of `Physics`.
"""
# Find a collision-free random initial configuration.
penetrating = True
while penetrating:
randomizers.randomize_limited_and_rotational_joints(physics, self.random)
# Check for collisions.
physics.after_reset()
penetrating = physics.data.ncon > 0
super().initialize_episode(physics)
def get_observation(self, physics):
"""Returns either the pure state or a set of egocentric features."""
obs = collections.OrderedDict()
if self._pure_state:
obs['position'] = physics.position()
obs['velocity'] = physics.velocity()
else:
obs['joint_angles'] = physics.joint_angles()
obs['head_height'] = physics.head_height()
obs['extremities'] = physics.extremities()
obs['torso_vertical'] = physics.torso_vertical_orientation()
obs['com_velocity'] = physics.center_of_mass_velocity()
obs['velocity'] = physics.velocity()
return obs
def get_reward(self, physics):
"""Returns a reward to the agent."""
standing = rewards.tolerance(physics.head_height(),
bounds=(_STAND_HEIGHT, float('inf')),
margin=_STAND_HEIGHT/4)
upright = rewards.tolerance(physics.torso_upright(),
bounds=(0.9, float('inf')), sigmoid='linear',
margin=1.9, value_at_margin=0)
stand_reward = standing * upright
small_control = rewards.tolerance(physics.control(), margin=1,
value_at_margin=0,
sigmoid='quadratic').mean()
small_control = (4 + small_control) / 5
if self._move_speed == 0:
horizontal_velocity = physics.center_of_mass_velocity()[[0, 1]]
dont_move = rewards.tolerance(horizontal_velocity, margin=2).mean()
return small_control * stand_reward * dont_move
else:
com_velocity = np.linalg.norm(physics.center_of_mass_velocity()[[0, 1]])
move = rewards.tolerance(com_velocity,
bounds=(self._move_speed, float('inf')),
margin=self._move_speed, value_at_margin=0,
sigmoid='linear')
move = (5*move + 1) / 6
return small_control * stand_reward * move