Berkeley Humanoid Lite
Open-source sub-$5K humanoid robot with 3D-printed gearboxes and full sim-to-real pipeline
License
MIT
Overview
Berkeley Humanoid Lite is a fully open-source humanoid robot platform designed to cost under $5,000. It features modular 3D-printed gearboxes using widely available components, with a complete software stack covering policy training in NVIDIA Isaac Lab, sim-to-sim validation in MuJoCo, and real-world deployment with motion capture and teleoperated manipulation. Developed by UC Berkeley's Hybrid Robotics Lab. The official build guide is mostly YouTube videos with a text BOM, firmware and calibration pages, and the actuator used for each joint is shown only in the videos.
Berkeley Humanoid Lite is a fully open-source humanoid robot that costs under $5,000 to build. It features modular 3D-printed gearboxes, complete CAD files, URDF/MJCF/USD robot descriptions, and a full software pipeline from Isaac Lab training through MuJoCo validation to real-world deployment.
Why it matters
Humanoid robots have traditionally cost $50K-$100K+, limiting research to well-funded labs. Berkeley Humanoid Lite democratizes humanoid research by providing a complete, affordable, and reproducible platform with both hardware designs and a full sim-to-real software pipeline from UC Berkeley.
Problem it solves
Access to humanoid robot hardware for research has been gatekept by prohibitive costs. This project provides a buildable, hackable humanoid platform that any university lab, maker, or researcher can construct for locomotion, manipulation, and whole-body control research.
Key takeaways
- Complete humanoid robot buildable for under $5,000
- 3D-printable gearboxes with modular joint design
- Full sim-to-real pipeline: Isaac Lab training, MuJoCo validation, real-world deployment
- Open CAD files, URDF, MJCF, and USD robot descriptions
Who it is for
- University robotics labs wanting affordable humanoid platforms
- Researchers studying humanoid locomotion and manipulation
- Advanced makers interested in building full-scale humanoid robots
Limitations
- Assembly requires significant 3D printing capacity and mechanical expertise
- Capabilities limited compared to commercial humanoids
- 3D-printed gearboxes may have durability concerns under heavy use
- Requires NVIDIA GPU for Isaac Lab training