UCLA researchers built a $400 humanoid using a $50 smartphone

A research team at UCLA's Robotics and Mechanisms Laboratory has published details of a bipedal humanoid robot that costs roughly $400 to build, anchored by a $50 off-the-shelf Android smartphone. The paper, posted to arXiv as the PhoneBot arXiv paper, describes a platform the researchers call PhoneBot that is designed to collapse the cost barrier separating academic robotics from real-world experimentation.

The cheapest comparable humanoid platform previously available are around $3,500: think the Flourish 1 humanoid that we profiled last month. 

PhoneBot undercuts that by a factor of eight to nine, while still delivering a capable mobile robot that can walk on flat ground, track and follow a moving person, and recover from falls without human intervention.

On the other hand, PhoneBot is only 43 centimeters tall (17 inches) and does not currently include arms or hands.

How a smartphone becomes a robot brain

The core insight behind PhoneBot is that modern smartphones already pack the sensors a bipedal robot needs most: an inertial measurement unit for balance, a camera for vision, a GPU-capable processor for inference, and wireless connectivity for remote operation. 

By mounting a standard Android handset at the robot's head, the team at UCLA's RoMeLa lab avoided purchasing dedicated sensor boards, compute modules, and communication hardware that together account for a large share of conventional robot costs.

The humanoid frame itself has 13 degrees of freedom spread across the legs, hips, and torso. Joints are driven by low-cost servo actuators selected to keep the bill of materials within budget while providing enough torque for stable locomotion. The full assembly prints or sources from commodity suppliers, and the researchers have released all design files, firmware, and software under an open-source license.

PhoneBot can walk, follow you, and recover from falls

In experiments described in the paper, PhoneBot demonstrated stable forward walking on level surfaces, real-time person-following using the phone's camera feed, and a fall-recovery routine that brings the robot back to a standing position after toppling. 

These three behaviors, basic as they may seem, represent the core functional requirements for a humanoid platform used in locomotion research.

The team notes that the robot is not intended to compete with commercial humanoids on payload, speed, or dexterity. Its purpose is to give university labs, high school programs, and independent researchers access to a bipedal testbed that does not require grant funding to acquire. 

The arXiv paper includes full build instructions alongside the source code.

Open-source timing matters

PhoneBot arrives at a moment when the humanoid robotics field is bifurcating sharply. 

On the one side you have well-funded commercial players spending tens of millions on hardware development; on the other a broader research community that largely cannot afford to participate. Platforms like Boston Dynamics' Spot and Figure's humanoids cost anywhere from tens to hundreds of thousands of dollars per unit, placing them out of reach for most academic groups.

The UCLA design directly addresses that gap. 

By publishing every component of the system, including the mechanical design, electronics layout, and the control software stack, the team is betting that a low-cost open platform will accelerate research contributions across institutions that would otherwise be locked out of bipedal locomotion work entirely.

The paper lists Dennis Hong, the founder of RoMeLa and a prominent figure in humanoid robotics research, as a senior author. Hong's lab previously developed CHARLI, one of the first full-size humanoid robots to compete in the RoboCup adult-size soccer league, and has consistently prioritized accessible hardware designs as a research philosophy.

Better soon?

One thing is likely here: people will play with PhoneBot, experiment with PhoneBot, and find ways to extend PhoneBot with better features and capabilities.

(Like arms and hands.)

And that will be super-helpful for robotics students everywhere.

Sources

PhoneBot arXiv preprint 2610.08737
UCLA RoMeLa lab official site