manumerous/wb_humanoid_mpc

Whole-Body Nonlinear MPC for Realtime Humanoid Loco-Manipulation Planning and Control

What it solves

This project provides a real-time, physics-based controller for humanoid robots, enabling them to perform complex movements like walking and manipulation tasks. It solves the challenge of managing the high-dimensional dynamics of a humanoid robot in real-time to generate stable, fluid behaviors.

How it works

The project implements a Nonlinear Model Predictive Controller (NMPC) that optimizes movements based on two different formulations:

  • Centroidal Dynamics MPC: Optimizes whole-body kinematics and center-of-mass dynamics, supporting either a single rigid body model or full centroidal dynamics with 6 DoF contact.
  • Whole-Body Dynamics MPC: Optimizes contact forces and joint accelerations, with the option to compute joint torques for every step in the planning horizon.

It is built upon an extended version of the ocs2 framework and utilizes libraries like pinocchio and hpipm for dynamics and optimization.

Who it’s for

Robotics researchers and developers working with humanoid platforms (such as the Unitree G1) who need a high-performance, torque-level control system for loco-manipulation.

Highlights

  • Real-time Torque Optimization: Directly optimizes full-order torque-level dynamics in real-time.
  • Hardware Agnostic: Designed to be platform-agnostic, though it supports examples for the Unitree G1 and 1X Neo.
  • Interactive Control: Supports base velocity and height control via Xbox gamepad or a GUI.
  • Dockerized Setup: Provides a Dockerfile and Dev Container support for easier installation and development.

Related

  • Project
    iit-DLSLab/Quadruped-PyMPC

    A Python-based model predictive controller for quadruped robots that offers both gradient-based and GPU-accelerated sampling-based optimization for stable locomotion.

  • Project
    leggedrobotics/ocs2

    A C++ toolbox for solving nonlinear optimal control problems, enabling real-time Model Predictive Control (MPC) for robotic systems with limited onboard computation.

  • Project
    google-deepmind/mujoco_mpc

    A software framework for real-time predictive control using MuJoCo, enabling the authoring and solving of complex robotics tasks through various planning methods.

  • Project
    hanruihua/RDA-planner

    An optimization-based MPC motion planner that uses ADMM to enable real-time, collision-free navigation for robots in cluttered environments.

  • Project
    acados/acados

    A high-performance C-based software package providing embedded solvers for nonlinear optimal control, designed for real-time Model Predictive Control and Moving Horizon Estimation.