mangdangroboticsclub/mini_pupper

Open-source physical AI platform bridging ROS and RL — from learning to earning.

What it solves

Mini Pupper 4 provides an open-source hardware and software platform designed to bridge the gap between simulated robotics training and real-world deployment. It allows developers to train complex neural network policies in high-fidelity environments and execute them on physical hardware.

How it works

The platform integrates high-performance 3-DoF coreless smart servos with a robust software stack. It utilizes MuJoCo and NVIDIA Isaac for physics simulation, incorporating Better Actuator Models (BAM) to account for joint friction. The system supports ROS 2 Jazzy for high-level autonomous navigation (SLAM, path planning) and micro-ROS for real-time, low-level actuator control on the onboard microcontroller.

Who it’s for

This project is for robotics researchers, students, and engineers interested in quadruped locomotion, reinforcement learning, and autonomous navigation. It is designed for those who want to move from theoretical simulation to physical robot implementation using a standardized, modular hardware base.

Highlights

  • Sim2Real Capability: Seamless transition from neural network training in MuJoCo/Isaac to real-world execution.
  • Advanced Actuator Modeling: Native integration with BAM to simulate precise joint friction and mechanical behavior.
  • Modern ROS 2 Support: Out-of-the-box compatibility with ROS 2 Jazzy for advanced autonomous navigation stacks.
  • Full-Stack Open Source: Includes everything from firmware and low-level micro-ROS drivers to high-level Raspberry Pi application code and mechanical blueprints.

Related

  • Project
    MIT-SPARK/Hydra

    A real-time spatial perception system that incrementally builds and optimizes 3D Scene Graphs for robotic environment understanding.

  • Project
    simpler-env/SimplerEnv

    A simulated evaluation framework for robot manipulation policies that uses visual matching and variant aggregation to provide a scalable, reproducible alternative to real-world testing.

  • Project
    hypery11/flipper-tesla-fsd

    A DIY toolkit for Flipper Zero (and ESP32) that interacts with Tesla's CAN bus to unlock FSD features, bypass nag warnings, restore banned features, and provide live diagnostics, as a free alternative to commercial devices.

  • Project
    lichtblick-suite/lichtblick

    Lichtblick is an integrated visualization and diagnosis tool for robotics that allows developers to inspect and visualize robotic data via a browser or desktop app.