Best Open-Source Robotics Projects 2026

The operating system of modern robotics. Every serious robot — from Boston Dynamics Spot to surgical systems — runs on ROS 2 or integrates with it. Not a single project but a meta-repo pointing to the full distro.

Hugging Face's end-to-end robot learning library. Load pretrained policies, fine-tune on your own robot arm data, and run inference in real-time. The fastest path from zero to a robot that learns from demonstrations.

GPU-accelerated robot learning on top of Isaac Sim. Run 4,096 parallel robot environments at 100,000 FPS for RL training. Trains policies for ANYmal, Franka, and humanoid robots in hours instead of days.

The reference navigation stack for ROS 2 mobile robots. Used in delivery robots, warehouse AMRs, and research platforms. Includes SLAM, costmap, behavior trees, and recoveries.

A curated collection of high-quality robot models for MuJoCo simulation — Franka, Spot, Shadow Hand, UR10, Unitree H1, and 50+ more. Ready to load, physics-accurate, with tuned contacts and actuators.

Turn a $50 RC car chassis and an old smartphone into an autonomous robot. OpenBot uses a phone as the compute brain (TensorFlow Lite inference) with an Arduino controlling motors.

Open-source humanoid robot with a Python SDK, ROS 2 integration, and a full simulation environment. Reachy 2 is commercially available but the SDK and sim are fully open — learn manipulation on a real humanoid design.

The go-to RL library for robotics research. Implements PPO, SAC, TD3, A2C with clean PyTorch code and extensive documentation. Pairs with MuJoCo, Gymnasium, and IsaacGym environments.

The default SLAM solution in ROS 2 Nav2. Builds 2D maps from lidar scans in real-time, supports lifelong mapping, and can serialize and deserialize maps across sessions.

The foundational computer vision library used in every robot with a camera. Object detection, depth estimation, pose estimation, and optical flow — all in one library with Python bindings.

Meta AI's Python abstraction layer for robotics. Provides a unified API for LoCoBot (navigation) and Sawyer (manipulation) — write one Python script, run on multiple robot platforms.

The next-generation Gazebo simulator. Physics-accurate simulation with lidar, cameras, IMU, and GPU rendering. The standard simulation environment for ROS 2 development.

Official SDK for Unitree Go2, B2, H1, and G1 robots. If you own or have access to a Unitree robot, this is the entry point for custom programming — from low-level joint control to high-level locomotion modes.

The standard interface for reinforcement learning environments. Every RL algorithm paper uses Gymnasium environments. Includes robot control tasks: Hopper, Ant, HalfCheetah, FetchReach, and more.

Run ROS 2 on microcontrollers — Arduino, ESP32, STM32, Raspberry Pi Pico. Enables your embedded firmware to publish sensor data and receive commands directly into the ROS 2 graph.