Berkeley Humanoid Lite: 3D-Printed Gearboxes Drive a $5K Humanoid
UC Berkeley’s open-source humanoid isn’t just a printed plastic body. The team is 3D printing the cycloidal gearboxes inside the robot’s actuators—the actual components that translate motor power into joint movement.
Why cycloidal gears?
The obvious objection to printing robot joints in plastic is strength. Humanoid joints take real loads. Berkeley’s answer is the cycloidal design: unlike standard spur gears with small teeth, cycloidal mechanisms distribute force across larger contact areas, which cuts down stress and wear on the printed parts.
The team has stress-tested these actuators extensively and reports they haven’t broken a single one during their experiments. For a printed mechanism, that’s a strong claim.
The build
- Height: ~1 meter
- Weight: ~16 kg (35 lb)
- Est. hardware cost: Under $5,000 (U.S. market pricing)
- Assembly time: A novice could potentially get one running in about a week
Each actuator module is a motor, a 3D-printed cycloidal gearbox, a position encoder, and standard bearings/hardware. The housing, gear, and shafts are all printed. You build the actuator, then you build the robot from them.
The repair loop is the real win
This is the part I’d actually get excited about if I were running a robotics lab. Robots fall. Parts wear. Gearboxes fail. With a proprietary system, a failed component is an expensive, complicated problem. With this one:
- Grab the open-source design.
- Print a new gear or shaft.
- Reassemble the actuator.
- Keep going.
The entire platform is designed around modularity and repairability. That’s a massive quality-of-life improvement for research environments where hardware is constantly being modified, tested, and occasionally destroyed.
The software is open, too
It’s not just CAD files. Berkeley has released the hardware designs, embedded code, training tools, and deployment framework. They’ve also demonstrated locomotion via reinforcement learning, including sim-to-real transfer. This is a research platform, not a kit.
What this means for the rest of us
The traditional hardware iteration loop is slow: design, send to a machine shop, wait, test, find the flaw, repeat. Desktop printing collapses that loop to: design, print, test, tweak, reprint. For anyone working on mechanisms, that feedback speed changes what’s feasible to prototype.
This isn’t a $5,000 replacement for a commercial household robot. It’s a serious research and education platform that lowers the barrier to humanoid robotics. And it’s a concrete example of 3D printing moving past decorative parts into functional, load-bearing mechanisms.
If you’re following this space, this is the project to watch.
