Robotics harnesses for AI agents
28 open-source Robotics harnesses an AI agent can use — MCP servers, SDKs, and adapters. Browse them on Loadbay. An agent can search these over Loadbay's MCP:
claude mcp add --transport http loadbay https://loadbay.xyz/api/mcp
→ Best Robotics harnesses (top picks, ranked)
- Genesis — Simulation platform for general-purpose robotics and embodied-AI learning with a unified physics engine.
- lerobot — End-to-end robot-learning library with pretrained policies, datasets, and sim environments for real-world robotics.
- PyBullet (Bullet3) — Bullet Physics SDK with the widely used PyBullet module for real-time simulation for robotics, RL, and sim-to-real.
- CARLA — Open-source simulator for autonomous-driving research with photorealistic urban environments, sensors, and a Python API.
- MuJoCo — General-purpose physics simulator, the de-facto standard engine for contact-rich robot learning and control.
- openpi — Physical Intelligence open release of pi0/pi0.5 vision-language-action models and runtime for robot manipulation.
- IsaacLab — Unified framework for robot learning built on NVIDIA Isaac Sim with GPU-accelerated simulation environments.
- Isaac GR00T — NVIDIA open foundation model for generalist humanoid robots, with fine-tuning and inference for vision-language-action control.
- OpenVLA — Open 7B vision-language-action model for robotic manipulation, mapping camera images plus language into robot actions.
- UniAD — Planning-oriented end-to-end autonomous-driving framework unifying perception, prediction, and planning.
- dm_control — DeepMind software stack for physics-based simulation and continuous-control RL environments built on MuJoCo.
- Mobile ALOHA — Stanford system for learning bimanual mobile manipulation via low-cost whole-body teleoperation and imitation learning.
- Drake — Toolbox for model-based design, simulation, and verification of robots, with multibody dynamics, optimization, and planning.
- ha-mcp — Home Assistant MCP server that lets AI agents query and control smart-home devices and automations.
- habitat-lab — Modular library to train embodied-AI agents for navigation and rearrangement across simulated environments.
- legged_gym — Isaac Gym RL environments for training quadruped and legged-robot locomotion policies.
- ManiSkill — GPU-parallelized robotics simulator and benchmark for manipulation skill learning built on SAPIEN.
- robosuite — Modular simulation framework and benchmark for robot learning built on the MuJoCo physics engine.
- Open X-Embodiment — DeepMind largest open real-robot dataset (RT-X) plus tooling to load and train cross-embodiment generalist robot policies.
- ROSA — ROSA, from NASA JPL: an agent that talks to ROS1/ROS2 robots in natural language, so you can command and inspect a robot stack by conversation.
- ros-mcp-server — MCP server connecting LLM agents to live robots through ROS and ROS2, exposing topics, services, and actions.
- Gymnasium-Robotics — Farama collection of gym-style robotics RL environments (Fetch, Shadow Hand, MaMuJoCo) with the standard Gymnasium API.
- homeassistant-mcp — MCP server exposing Home Assistant entities and services so AI assistants can operate a smart home.
- rai — A vendor-agnostic agentic framework for physical-AI robotics on ROS 2, wiring LLM agents into real robot control.
- hass-mcp — Home Assistant MCP server for controlling smart-home devices through AI assistants via the Model Context Protocol.
- embodied-agents — Toolkit for integrating state-of-the-art transformer models into robotics stacks for embodied agents.
- rosclaw — Infrastructure that grounds AI agents in the physical world via e-URDF, sandbox safety, memory, and self-evolving skills.
- embodied-agents — ROS2-based framework for building interactive physical agents that understand, remember, and act on their environment.