google-deepmind/mujoco_mpc

Real-time behaviour synthesis with MuJoCo, using Predictive Control

What it solves

MuJoCo MPC (MJPC) provides a framework for real-time predictive control of robots. It allows users to author and solve complex robotics tasks by predicting future states and optimizing actions in real-time, reducing the manual effort required to program complex behaviors.

How it works

MJPC integrates with the MuJoCo physics engine to implement multiple shooting-based planners. It offers both derivative-based optimization methods, such as iLQG and Gradient Descent, and derivative-free methods, specifically a "Predictive Sampling" planner. The system includes a graphical user interface (GUI) for interaction and an experimental Python API for programmatic control.

Who it’s for

This tool is designed for robotics researchers and developers who need to implement and test real-time control strategies for complex robotic systems, such as quadrupeds, humanoid robots, or bimanual manipulators.

Highlights

  • Supports multiple planning methods including iLQG, Gradient Descent, and Predictive Sampling.
  • Includes a dedicated GUI for interactive task authoring and solving.
  • Capable of handling diverse tasks like Rubik's cube unscrambling and motion-capture tracking.
  • Provides a Python API for integrating MJPC with other Python-based workflows.

Related

  • Project
    ARISE-Initiative/robosuite

    A MuJoCo-powered simulation framework and benchmark suite for robot learning, providing modular tools to create environments and test AI control algorithms.

  • 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
    google-deepmind/mujoco_menagerie

    A curated collection of high-quality robot models for the MuJoCo physics engine, providing reliable and well-designed simulations for a wide range of robotic platforms.

  • Project
    manumerous/wb_humanoid_mpc

    A real-time Nonlinear Model Predictive Controller (NMPC) for humanoid robots that optimizes full-order torque-level dynamics for loco-manipulation.

  • Project
    google-deepmind/mujoco

    A high-performance physics engine for the fast and accurate simulation of articulated structures, widely used in robotics and machine learning research.