mujoco-py Humanoid
mujoco-py Humanoid
table: observation: Box(376)
Num Observation Min Max Mean
0 root_z (Position) -inf +inf 1.4
1 root_w (global orientation quaternion) -1 +1 1
2 root_x (global orientation quaternion) -1 +1 0
3 root_y (global orientation quaternion) -1 +1 0
4 root_z (global orientation quaternion) -1 +1 0
5 abdomen_z (rotation) -45 +45 0
6 abdomen_y (rotation) -75 +30 0
7 abdomen_x (rotation) -35 +35 0
8 right_hip_x (rotation) -25 +5 0
9 right_hip_z (rotation) -60 +35 0
10 right_hip_y (rotation) -110 +20 0
11 right_knee (rotation) -160 -2 0
12 left_hip_x (rotation) -25 +5 0
13 left_hip_z (rotation) -60 +35 0
14 left_hip_y (rotation) -110 +20 0
15 left_knee (rotation) -160 -2 0
16 right_shoulder_1 (rotation) -85 +60 0
17 right_shoulder_2 (rotation) -85 +60 0
18 right_elbow (rotation) -90 +50 0
19 left_shoulder_1 (rotation) -85 +60 0
20 left_shoulder_2 (rotation) -85 +60 0
21 left_elbow (rotation) -90 +50 0
22:45 qvel (velocity of joints) unknown unknown unknown
45:185 cinert (center of mass based inertia) unknown unknown unknown
185:269 cvel (center of mass based velocity) unknown unknown unknown
269:292 qfrac_actuator (actuator force) unknown unknown unknown
292:376 cfrac_ext (center of mass based external force) unknown unknown unknown
0:21 : qpos
The openai gym environment hides first 2 dimensions of qpos returned by MoJoCo.
The correspond to x and y co-ordinate of the robot root (abdomen).
The reason is these quantity can grow boundlessly and their absolute value do not carry any significance.
If you want to obtain the full state just use env.env.state_vector() and you will get a 47 dimensional vector containing qpos and qvel. reffer to mujoco_env.py.
All the joint angles are in radians.
shoulder_1 swings hand parallel to the direction in which the body is facing.
table: action Box(17)
Num Name Gear ObservationでのNum
0 abdomen_y 100 6
1 abdomen_z 100 5
2 abdomen_x 100 7
3 right_hip_x 100 8
4 right_hip_z 100 9
5 right_hip_y 300 10
6 right_knee 200 11
7 left_hip_x 100 12
8 left_hip_z 100 13
9 left_hip_y 300 14
10 left_knee 200 15
11 right_shoulder1 25 16
12 right_shoulder2 25 17
13 right_elbow 25 18
14 left_shoulder1 25 19
15 left_shoulder2 25 20
16 left_elbow 25 21
Torque control