Implementation:Haosulab ManiSkill Humanoid
| Knowledge Sources | |
|---|---|
| Domains | Robotics, Simulation, Humanoid_Control |
| Last Updated | 2026-02-15 08:00 GMT |
Overview
Concrete tool for simulating the MuJoCo-style Humanoid robot in ManiSkill environments.
Description
The Humanoid is a 21-DOF bipedal robot loaded from an MJCF (MuJoCo XML) file. It features a free root link (fix_root_link=False), an abdomen with 3 joints (x, y, z), two legs each with 5 joints (hip x/y/z, knee, ankle x/y), and two arms each with 3 joints (shoulder1, shoulder2, elbow). The pd_joint_delta_pos controller uses per-joint stiffness and damping values tuned for different body parts: higher stiffness for hips (40-120) and knees (80) versus lower for shoulders (20) and elbows (40). Gravity compensation is disabled (balance_passive_force=False) since the robot is free-standing.
uid: humanoid
MJCF path: {PACKAGE_ASSET_DIR}/robots/humanoid/humanoid.xml
Supported control modes: pd_joint_pos, pd_joint_delta_pos
Usage
Use Humanoid for locomotion and whole-body control tasks such as standing, walking, running, and acrobatic movements. The per-joint tuned controller parameters make it suitable for learning complex locomotion policies. Based on the classic MuJoCo humanoid model.
Code Reference
Source Location
- Repository: Haosulab_ManiSkill
- File: mani_skill/agents/robots/humanoid/humanoid.py
Signature
@register_agent()
class Humanoid(BaseAgent):
uid = "humanoid"
mjcf_path = f"{PACKAGE_ASSET_DIR}/robots/humanoid/humanoid.xml"
fix_root_link = False
keyframes = dict(squat=Keyframe(qpos=np.array([...]), pose=sapien.Pose(p=[0, 0, 1.13])))
Import
from mani_skill.agents.robots.humanoid.humanoid import Humanoid
I/O Contract
Inputs
| Name | Type | Required | Description |
|---|---|---|---|
| (inherited from BaseAgent) |
Outputs
| Name | Type | Description |
|---|---|---|
| robot agent | Humanoid | Configured 21-DOF bipedal humanoid with per-joint tuned PD controllers |
Usage Examples
Creating Environment with Robot
import gymnasium as gym
import mani_skill.envs
env = gym.make("HumanoidStandup-v1", robot_uids="humanoid")
obs, info = env.reset()