-
Notifications
You must be signed in to change notification settings - Fork 2
/
pusher.py
96 lines (77 loc) · 3.02 KB
/
pusher.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
from __future__ import division
from __future__ import print_function
from __future__ import absolute_import
import os
import numpy as np
from gym import utils
from gym.envs.mujoco import mujoco_env
class PusherEnv(mujoco_env.MujocoEnv, utils.EzPickle):
def __init__(self):
dir_path = os.path.dirname(os.path.realpath(__file__))
mujoco_env.MujocoEnv.__init__(self, '%s/assets/pusher.xml' % dir_path, 4)
utils.EzPickle.__init__(self)
self.reset()
def step(self, a):
obj_pos = self.get_body_com("object"),
vec_1 = obj_pos - self.get_body_com("tips_arm")
vec_2 = obj_pos - self.get_body_com("goal")
reward_near = -np.sum(np.abs(vec_1))
reward_dist = -np.sum(np.abs(vec_2))
reward_ctrl = -np.square(a).sum()
reward = 1.25 * reward_dist + 0.1 * reward_ctrl + 0.5 * reward_near
self.do_simulation(a, self.frame_skip)
ob = self._get_obs()
done = False
return ob, reward, done, {}
def get_true_pos(self, s, a):
obj_pos = s[-3:],
vec_1 = obj_pos - s[-6:-3]
vec_2 = obj_pos - self.get_body_com("goal")
reward_near = -np.sum(np.abs(vec_1))
reward_dist = -np.sum(np.abs(vec_2))
reward_ctrl = -np.square(a).sum()
reward = 1.25 * reward_dist + 0.1 * reward_ctrl + 0.5 * reward_near
self.do_simulation(a, self.frame_skip)
ob = self._get_obs()
done = False
return ob
def _step(self, a):
obj_pos = self.get_body_com("object"),
vec_1 = obj_pos - self.get_body_com("tips_arm")
vec_2 = obj_pos - self.get_body_com("goal")
reward_near = -np.sum(np.abs(vec_1))
reward_dist = -np.sum(np.abs(vec_2))
reward_ctrl = -np.square(a).sum()
reward = 1.25 * reward_dist + 0.1 * reward_ctrl + 0.5 * reward_near
self.do_simulation(a, self.frame_skip)
ob = self._get_obs()
done = False
return ob, reward, done, {}
def viewer_setup(self):
self.viewer.cam.trackbodyid = -1
self.viewer.cam.distance = 4.0
def reset(self):
qpos = self.init_qpos
self.goal_pos = np.asarray([0, 0])
self.cylinder_pos = np.array([-0.25, 0.15]) + np.random.normal(0, 0.025, [2])
qpos[-4:-2] = self.cylinder_pos
qpos[-2:] = self.goal_pos
qvel = self.init_qvel + self.np_random.uniform(low=-0.005,high=0.005, size=self.model.nv)
qvel[-4:] = 0
self.set_state(qpos, qvel)
self.ac_goal_pos = self.get_body_com("goal")
return self._get_obs()
def _get_obs(self):
return np.concatenate([
self.model.data.qpos.flat[:7],
self.model.data.qvel.flat[:7],
self.get_body_com("tips_arm"),
self.get_body_com("object"),
])
def get_obs(self):
return np.concatenate([
self.model.data.qpos.flat[:7],
self.model.data.qvel.flat[:7],
self.get_body_com("tips_arm"),
self.get_body_com("object"),
])