BERKELEY OPEN-SOURCED A HUMANOID ROBOT YOU CAN 3D PRINT FOR A FIFTH OF THE PRICE
UC Berkeley published complete plans for a 90 cm humanoid that walks, manipulates objects, and costs a twentieth of a commercial one.
by UC Berkeley Hybrid Robotics Lab
RoboticsOpen-hardware
- difficulty
- ●●●●●
- time
- months
- license
- MIT
- repo
- repo ACTIVE1,450 stars
●●●●● · months · MIT · 1,450 stars · repo ACTIVE
WHAT YOU’LL NEED
- 3D printer + filament — printable parts — files are in the repo
- Electronic parts — full list with part numbers in the repo BOM
- Dev board / microcontroller — runs the project firmware
- Soldering kit + basic tools — one-time buy, serves every future build
Partner
COMPAREE VERDICT
This is the most complete open humanoid project published by an academic lab. The mechanical design is elegant: 3D-printed cycloidal gearboxes replace the 1,000 dollars+ harmonic drives that make most humanoids unaffordable, and the BOM is entirely off-the-shelf motors, encoders, and structure. It walks, balances, picks up objects, solves a Rubik's cube, and writes with a marker — all skills trained via reinforcement learning in simulation then transferred to hardware. The repository includes every file you need: CAD, STLs, firmware, simulation environment, and training code. The cost claim of under 5,000 dollars is real if you can print the parts yourself and source carefully; commercial quotes will push it higher. The single thing most likely to go wrong is the cycloidal gearboxes. They are the entire reason this project is affordable, but they are also unforgiving: print quality, bearing press-fit tolerances, and backlash tuning all matter, and there is no step-by-step tutorial for assembly or troubleshooting. If you have built printed gearboxes before or are prepared to iterate, this is a landmark project. If you have not, expect to reprint parts and learn as you go. This is not a weekend kit — it is a months-long build for someone with machining access, electronics experience, and patience.
IN THE REPO
NOT IN THE REPO
- —Full mechanical CAD (STEP and Fusion 360), STL files for every printed part, and assembly drawings are in the repository.
- —Bill of materials includes part numbers for motors, bearings, encoders, and fasteners — most from McMaster-Carr and AliExpress.
- —Firmware and control software are present, plus simulation code and reinforcement learning training scripts.
- —Assembly guide is detailed but assumes experience with cycloidal gearboxes, motor wiring, and inverse kinematics.
- —MIT license: you can build, modify, and sell this. The repository is actively maintained.
- —No official video tutorial — the UC Berkeley lab video shows the robot in action but not the build process.
Can I build this?
Build at your own risk. Projects involve tools, electronics and sometimes mains voltage — follow the creator’s safety notes.
Partner · KickstarterHeyGears G1: 10M+ colors and transparent parts in one print, plus UV printing on flat objects. Figures, parts and labels — no painting.
Videos
Berkeley Humanoid Lite: An Open source, Accessible, and Customizable 3D printed Humanoid Robot.
No official build tutorial exists. The UC Berkeley Hybrid Robotics lab has published a video showing the robot walking and manipulating objects, but it is a demonstration, not a how-to. Community walkthroughs may appear as people build this.
Gallery
Start here
Navigation into the creator’s own docs — we don’t rewrite the guide, we route you to the source.
- 1.Read the README and the assembly guide PDF in the repository. (The README links to the mechanical files, BOM, and firmware. The assembly guide is in the docs folder.)
- 2.Review the BOM and decide whether to source from AliExpress or domestic suppliers.(The BOM includes McMaster-Carr part numbers (US) and generic equivalents. Shipping and lead time vary widely.)
- 3.Print a single gearbox assembly first and test fit before committing to the full skeleton.(The cycloidal gearboxes are the hardest part. If your printer cannot hold the tolerances, you will know immediately.)
- 4.Set up the simulation environment to understand the control stack before you build hardware.(The repository includes MuJoCo simulation files and training scripts. Running these will clarify what the firmware expects.)
Resources
Documentation, files and community threads for this build — we link straight to the original sources and never rehost the creator’s files.
KNOWN ISSUES
- The BOM does not include a battery or power supply — you will need to spec one based on the motor current draw, which is not called out in a single place.
- Cycloidal gearboxes are sensitive to print quality and bearing press-fit. If your printer drifts or your bearings are loose, the backlash will make the robot unusable.
- The assembly guide assumes you know how to wire motor drivers and encoders. There is no wiring diagram for beginners.
- The reinforcement learning pipeline requires a GPU and familiarity with Python simulation frameworks. You cannot skip this and expect the robot to work out of the box.
- No official support forum or Discord. You are on your own unless the community builds one.
- The $5,000 cost assumes you print everything yourself and source aggressively. If you pay for prints or buy domestically, expect $6,000-7,000.
Can I build this without prior robotics experience?
No. The repository assumes you understand motor control, inverse kinematics, and tolerance stacking. If you have not built a robot with custom gearboxes before, this will be a steep learning curve.
Do I need access to a machine shop?
Not strictly, but it helps. The cycloidal gears are 3D printed, but if your printer cannot hold 0.1 mm tolerances consistently, you may need to bore or turn some parts.
Is the $5,000 cost realistic?
Yes, if you print at home and source from AliExpress. Add $1,000-2,000 if you pay for prints or buy from McMaster-Carr exclusively.
Can I use different motors?
Maybe. The control code expects specific motor torque and encoder resolution. Swapping motors means retuning the entire stack.
Is there a pre-trained model I can load?
The repository includes trained policies for walking and manipulation, but they are tuned to the hardware. You will still need to run the simulation and verify before deploying to hardware.
Community builds
No community builds yet — be the first, we feature the best ones.
Discussion1
FROM THE COMPAREE TEAM
The cycloidal gearboxes are what make this affordable, but they are also the hardest part to get right. If you have printed gearboxes before, what tolerances did you hold and what material worked best?
UC Berkeley Hybrid Robotics Lab
The Hybrid Robotics group at UC Berkeley focuses on legged robots and manipulation. They published Humanoid Lite to make humanoid research accessible outside of corporate labs, where robot platforms cost six figures and are rarely shared.
DISCLAIMER
- Comparee is not the author of the projects featured here. All rights to each project belong to its creator — every page links to the original source, and we never host creators’ files.
- Information is provided without warranty and may become outdated as projects evolve. Prices are indicative bands only — always check the creator’s parts list for current costs.
- Building and operating any project is at your own responsibility. Protective equipment, safe workshop practice and compliance with local regulations are the builder’s responsibility.


CompareeTEAM1mo agoedited
Practical notes from our verification: The repository is complete — CAD, BOM, firmware, and simulation are all present, and the MIT license is clean. The $5,000 cost is real if you source carefully and print at home, but it does not include a battery or power supply, and those are not specced in the BOM. The single biggest variable is the cycloidal gearboxes: if your printer holds tolerances, this is achievable; if not, you will reprint parts or need machining help. No official video tutorial exists — the lab video is a demo, not a build guide. This is the most complete open humanoid we have seen from an academic lab, but it is a months-long expert build, not a weekend kit.