iit-DLSLab/Quadruped-PyMPC

A model predictive controller for quadruped robots based on the single rigid body model and written in python. Gradient-based (acados) or Sampling-based (jax)

What it solves

It provides a high-performance model predictive controller (MPC) for quadruped robots, enabling them to perform complex movements like trotting, pacing, and bounding while maintaining stability and robustness.

How it works

The project implements a controller based on a single rigid body model using Python. It offers two distinct computational approaches:

  1. Gradient-based MPC: Uses acados for fast computation (under 5ms on CPU), featuring optional foothold optimization, ground reaction force smoothing, and stability criteria.
  2. Sampling-based MPC: Uses jax to run thousands of parallel rollouts on a GPU (under 2ms), supporting strategies like MPPI and CEMPPI with various control parametrizations (zero-order, linear, or cubic splines).

Who it’s for

It is designed for robotics researchers and engineers working on legged locomotion who need a flexible, Python-based MPC framework compatible with Mujoco and real-world hardware.

Highlights

  • Dual implementation supporting both gradient-based (CPU) and sampling-based (GPU) optimization.
  • Extremely low latency: <5ms for gradient-based and <2ms for sampling-based rollouts.
  • Supports advanced control strategies including MPPI, CEMPPI, and Feedback-MPPI.
  • Compatible with Mujoco simulation and real-world deployment via Unitree robot communication tools.

Related

  • 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
    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
    hanruihua/RDA-planner

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

  • Dispatch
    Anthropic Claude Robotics Evaluation Shows Rapid Gains on High‑Level Control but Limited Direct Torque Mastery

    Anthropic’s July 2026 study finds that newer Claude models can reliably supervise pretrained locomotion and manipulation policies, while direct low‑level motor control remains unreliable across robot bodies.

  • 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.