Berkeley Humanoid Lite Is a Sub-$5,000 Open Source Robot You Can 3D Print

Humanoid robots usually come with price tags that only well-funded labs can afford, and the designs are often locked behind NDAs or proprietary control boxes. A team at UC Berkeley is trying to change that with Berkeley Humanoid Lite, a full bipedal robot that costs less than $5,000 and can be built using standard desktop 3D printers and off-the-shelf parts.

Built around 3D-printed gearboxes

The robot stands 0.8 meters tall and weighs 16 kilograms. Its key innovation is a modular cycloidal gearbox for each actuator. The researchers chose this design because it offers good torque density in a compact package that works well with 3D-printed parts. Cycloidal drives do not need the same precision as planetary gears, which makes them easier to print reliably on a desktop machine.

Most structural parts are 3D printed. The motors, bearings, and electronics come from common suppliers like Mouser, DigiKey, and AliExpress-style vendors. According to the project page, every part needed for the robot is available online without custom machining or special fabrication equipment.

Open source all the way down

The Berkeley team released the hardware files, embedded code, simulation models, and reinforcement-learning training pipeline under an open-source license. The full repository is on GitHub, and the documentation is hosted on GitBook. The paper is available on arXiv, and the project website links to a YouTube video showing the robot walking, manipulating objects, writing its name, and even playing with a Rubik's cube.

This matters because most humanoid platforms either cost tens or hundreds of thousands of dollars, or they expose only a narrow API. Berkeley Humanoid Lite gives researchers, students, and hobbyists a complete platform they can modify, repair, and study from the ground up.

Does it actually work?

The researchers tested the actuators extensively to address the obvious concern: 3D-printed plastic gearboxes are weaker than metal ones. They then trained a locomotion controller with reinforcement learning and demonstrated zero-shot transfer from simulation to the real hardware. In plain terms, the policy that worked in simulation also worked on the physical robot without re-tuning.

The robot performed bipedal locomotion and teleoperated manipulation tasks in the project video. It is not ready to replace a Boston Dynamics machine, but it does not need to. The goal is to provide a capable, low-cost research platform, not a commercial product.

Why this could shift the field

Most robotics breakthroughs come from groups with expensive hardware stacks. That creates a bottleneck: only a few institutions can reproduce the results or build on them. By dropping the cost below $5,000 and making the design printable at home, Berkeley Humanoid Lite could let far more people contribute to humanoid control, locomotion, and manipulation research.

The project also sits at an interesting intersection between 3D printing and robotics. Desktop printers have long been used for brackets and prototypes. Here, they are producing structural mechanical components that must survive dynamic walking loads. If the actuators hold up over long use, it will be a strong signal that 3D-printed gearboxes can move beyond static parts and into moving machinery.

The bottom line

Berkeley Humanoid Lite is not the flashiest humanoid robot announcement of 2026, but it may be one of the most useful. It combines a low price, open-source documentation, and a design that can actually be replicated without a machine shop. For anyone interested in real humanoid robotics research on a budget, this is a project worth watching closely.