Chestnut-Robotics/aero-hand-open

https://shop.tetheria.ai/

What it solves

Aero Hand Open provides an affordable, accessible, and open-source alternative to expensive proprietary robotic hands. It enables researchers and enthusiasts to conduct dexterous manipulation learning without the high cost of industrial hardware.

How it works

The project is a 7-degree-of-freedom (DoF) robotic hand with 5 fingers and 16 joints. It uses a tendon-driven architecture to achieve smooth, natural motion. The physical structure is fully 3D-printed and uses off-the-shelf electronic components, making it easy to assemble and maintain. For control and training, it includes a Python SDK, ROS2 integration, and high-fidelity MuJoCo simulation models, along with reinforcement learning tools built on mujoco_playground to train custom policies.

Who it’s for

  • Robotics researchers focusing on dexterous manipulation.
  • Educational institutions and students.
  • Robotics enthusiasts and makers.

Highlights

  • Low Cost: A complete kit costs approximately $314 USD.
  • Lightweight: The entire hand weighs only 389 g.
  • Simulation-Ready: Includes MuJoCo models for high-fidelity simulation and RL training.
  • Open Ecosystem: Provides open-source hardware, firmware, and a Python SDK compatible with ESP32 and ROS2.

Related

  • Project
    AgileRL/AgileRL

    AgileRL is a Python library for deep reinforcement learning that adds *RLOps*: evolutionary hyper‑parameter optimisation, distributed training, and a cloud platform (Arena) for running experiments at scale. It ships many algorithms (PPO, DQN, TD3, MADDPG, bandits, LLM fine‑tuning, etc.) and lets you train a single agent or a population that evolves its hyper‑parameters automatically, promising up to 10× faster HPO than traditional methods.

  • Project
    Motphys/UniLab

    A heterogeneous architecture for robot reinforcement learning that decouples task definitions from physics simulators and hardware, enabling a seamless switch between backends like MuJoCo and IsaacGym.

  • Project
    nicklashansen/tdmpc2

    A scalable, robust model-based reinforcement learning algorithm for continuous control that can handle multiple tasks across different domains and embodiments using a single agent.

  • Project
    Farama-Foundation/MAgent2

    A library for creating pixel-based gridworld environments where large numbers of agents interact in competitive scenarios to study collective intelligence.