-
Notifications
You must be signed in to change notification settings - Fork 123
/
multi_ant.py
299 lines (242 loc) · 10.5 KB
/
multi_ant.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
import copy
import math
import sys
import os
from lxml import etree
import lxml.builder
import gym
import numpy as np
from gym import spaces
from gym import utils
from gym.utils import seeding
from gym.envs.mujoco import mujoco_env
from madrl_environments import AbstractMAEnv, Agent
from rltools.util import EzPickle
class AntLeg(Agent):
def __init__(self, model, idx, n_legs,
pos_noise=1e-3,
vel_noise=1e-3,
force_noise=1e-3):
self._idx = idx
self.n_legs = n_legs
self.model = model
self.pos_noise = pos_noise
self.vel_noise = vel_noise
self.force_noise = force_noise
def _seed(self, seed=None):
self.np_random, seed = seeding.np_random(seed)
return [seed]
@property
def action_space(self):
return spaces.Box(low=-1, high=1, shape=(2,))
@property
def observation_space(self):
# 18 force observations for each leg + 4 pos + vel per leg (2 neighboring legs) + 11 world coords
return spaces.Box(low=-np.inf, high=np.inf, shape=(18 + 4 + 4 + 4 + 11,))
def get_observation(self):
idx = self._idx
n1_idx = idx-1 if idx > 0 else self.n_legs-1
n2_idx = idx+1 if idx < (self.n_legs - 1) else 0
return np.concatenate([
np.random.normal(self.model.data.qpos.flat[2:7], self.pos_noise), # body pos
np.random.normal(self.model.data.qvel.flat[:6], self.vel_noise), # body vel
np.random.normal(self.model.data.qpos.flat[7+2*idx:9+2*idx], self.pos_noise),
np.random.normal(self.model.data.qvel.flat[6+2*idx:8+2*idx], self.vel_noise),
np.random.normal(self.model.data.qpos.flat[7+2*n1_idx:9+2*n1_idx], self.pos_noise),
np.random.normal(self.model.data.qvel.flat[6+2*n1_idx:8+2*n1_idx], self.vel_noise),
np.random.normal(self.model.data.qpos.flat[7+2*n2_idx:9+2*n2_idx], self.pos_noise),
np.random.normal(self.model.data.qvel.flat[6+2*n2_idx:8+2*n2_idx], self.vel_noise),
np.random.normal(np.clip(self.model.data.cfrc_ext[3*idx+2:3*idx+5], -1, 1).flat, self.force_noise)
])
class MultiAnt(EzPickle, mujoco_env.MujocoEnv):
def __init__(self,
n_legs=4,
ts=0.02,
integrator='RK4',
leg_length=0.282,
out_file="multi_ant.xml",
base_file="ant_og.xml",
reward_mech='local',
pos_noise=1e-3,
vel_noise=1e-3,
force_noise=1e-3
):
EzPickle.__init__(self, n_legs, ts, integrator, leg_length,
out_file, base_file, reward_mech,
pos_noise, vel_noise, force_noise)
self.n_legs = n_legs
self.ts = ts
self.integrator = integrator
self.leg_length = leg_length
self.out_file = out_file
self.base_file = base_file
self._reward_mech = reward_mech
self.pos_noise = pos_noise
self.vel_noise = vel_noise
self.force_noise = force_noise
self.legs = None
self.out_file_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), self.out_file)
self.base_file_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), self.base_file)
self.gen_xml(out_file=self.out_file_path, og_file=self.base_file_path)
mujoco_env.MujocoEnv.__init__(self, self.out_file_path, 5)
self.legs = [AntLeg(self.model, i, n_legs, pos_noise=pos_noise, vel_noise=vel_noise, force_noise=force_noise) for i in range(self.n_legs)]
@property
def agents(self):
return self.legs
@property
def reward_mech(self):
return self._reward_mech
def seed(self, seed=None):
self.np_random, seed_ = seeding.np_random(seed)
return [seed_]
def setup(self):
self.seed()
self.gen_xml(out_file=self.out_file_path, og_file=self.base_file_path)
mujoco_env.MujocoEnv.__init__(self, self.out_file_path, 5)
self.legs = [AntLeg(self.model, i, self.n_legs, pos_noise=self.pos_noise, vel_noise=self.vel_noise,
force_noise=self.force_noise) for i in range(self.n_legs)]
def _step(self, a):
xposbefore = self.get_body_com("torso")[0]
self.do_simulation(a, self.frame_skip)
xposafter = self.get_body_com("torso")[0]
forward_reward = (xposafter - xposbefore)/self.dt
ctrl_cost = .5 * np.square(a).sum()
contact_cost = 0.5 * 1e-3 * np.sum(
np.square(np.clip(self.model.data.cfrc_ext, -1, 1)))
survive_reward = 1.0
reward = forward_reward - ctrl_cost - contact_cost + survive_reward
state = self.state_vector()
notdone = np.isfinite(state).all() \
and state[2] >= 0.26 and state[2] <= 1.0
done = not notdone
ob = self._get_obs()
return ob, [reward]*self.n_legs, done, dict(
reward_forward=forward_reward,
reward_ctrl=-ctrl_cost,
reward_contact=-contact_cost,
reward_survive=survive_reward)
def _get_obs(self):
obs = [l.get_observation() for l in self.legs] if self.legs else np.random.rand(self.n_legs, 32)
return np.array(obs)
def reset_model(self):
qpos = self.init_qpos + self.np_random.uniform(size=self.model.nq,low=-.1,high=.1)
qvel = self.init_qvel + self.np_random.randn(self.model.nv) * .1
self.set_state(qpos, qvel)
return self._get_obs()
def viewer_setup(self):
self.viewer.cam.distance = self.model.stat.extent * 0.5
def gen_xml(self, out_file="ant.xml",
og_file="ant_og.xml"):
"""Write .xml file for the ant problem.
Modify original 4 leg ant defintion.
"""
self.init_geometry()
parser = etree.XMLParser(remove_blank_text=True)
og = etree.parse(og_file, parser)
# add legs
torso = og.find('.//body')
# first remove original legs
for c in torso.getchildren():
if c.tag == 'body':
torso.remove(c)
for i in range(self.n_legs):
# 1st half leg
leg = etree.SubElement(torso, "body",
name="leg_"+str(i),
pos="0 0 0")
etree.SubElement(leg, "geom",
name="aux_"+str(i)+"_geom",
type="capsule",
size="0.08",
fromto="0.0 0.0 0.0 "+self.leg_geom_string[i])
# Hip joint
aux = etree.SubElement(leg, "body",
name="aux_"+str(i),
pos=self.leg_geom_string[i])
etree.SubElement(aux, "joint",
name="hip_"+str(i),
type="hinge",
pos="0.0 0.0 0.0",
axis="0 0 1",
range="-30 30")
etree.SubElement(aux, "geom",
name="leg_geom_"+str(i),
type="capsule",
size="0.08",
fromto="0.0 0.0 0.0 "+self.leg_geom_string[i])
# ankle joint
ankle = etree.SubElement(aux, "body",
pos=self.leg_geom_string[i])
etree.SubElement(ankle, "joint",
name="ankle_"+str(i),
type="hinge",
pos="0.0 0.0 0.0",
axis=self.leg_axis[i],
range=self.angle_range[i])
etree.SubElement(ankle, "geom",
name="ankle_geom_"+str(i),
type="capsule",
size="0.08",
fromto="0.0 0.0 0.0 "+self.leg_geom_string_2x[i])
# add new motors
actuators = og.find('actuator')
actuators.clear()
for i in range(self.n_legs):
etree.SubElement(actuators, "motor",
joint="hip_"+str(i),
ctrlrange="-150.0 150.0",
ctrllimited="true")
etree.SubElement(actuators, "motor",
joint="ankle_"+str(i),
ctrlrange="-150.0 150.0",
ctrllimited="true")
og.write(out_file, pretty_print=True)
def init_geometry(self):
self.gen_leg_geometry()
def gen_leg_geometry(self):
self.leg_geom = np.zeros((self.n_legs, 3))
self.leg_geom_string = []
self.leg_geom_string_2x = []
self.leg_axis = []
self.angle_range = []
for i in range(self.n_legs):
x, y = np.round(self.get_point_on_circle(self.leg_length, i, self.n_legs), decimals=5)
self.leg_geom[i,:] = np.array([x,y,0.0])
self.leg_geom_string.append(str(x) + " " + str(y) + " " + "0.0")
self.leg_geom_string_2x.append(str(2*x) + " " + str(2*y) + " " + "0.0")
# TODO (max): y no switch?
if (x >= 0.0 and y >= 0.0):
self.angle_range.append("30 70")
self.leg_axis.append("-1 1 0")
elif (x > 0.0 and y < 0.0):
self.angle_range.append("30 70")
self.leg_axis.append("1 1 0")
elif (x <= 0.0 and y <= 0.0):
self.angle_range.append("-70 -30")
self.leg_axis.append("-1 1 0")
else:
self.angle_range.append("-70 -30")
self.leg_axis.append("1 1 0")
def get_point_on_circle(self, r, current_point, total_points):
theta = 2*np.pi / total_points
angle = theta * current_point + np.pi/self.n_legs
x = r * np.cos(angle)
y = r * np.sin(angle)
return x, y
def set_param_values(self, lut):
for k, v in lut.items():
setattr(self, k, v)
self.setup()
def get_param_values(self):
return self.__dict__
if __name__ == '__main__':
env = MultiAnt(4)
env.reset()
for i in range(250):
env.render()
a = np.array([l.action_space.sample() for l in env.agents])
o, r, done, _ = env.step(a)
print("\nStep:", i)
print("Rewards:", r)
if done:
break