UCLA Researchers Build $400 Humanoid Robot Powered by Android Smartphones

Modern smartphones combine high-resolution cameras, inertial sensors, and multi-core processors into a single self-contained unit. Aiming to reduce the hardware costs and assembly hurdles of robotics research, engineers at the University of California, Los Angeles have designed PhoneBot, a small bipedal humanoid robot that mounts an Android smartphone as its primary perception engine and locomotion controller.

According to research detailed in an arXiv preprint and reported by TechXplore (https://techxplore.com/news/2026-10-phonebot-smartphones-job-legs.html), the robot stands approximately 19 inches (48 centimeters) tall and weighs around 4 pounds (1.8 kilograms) with a phone attached. Excluding the smartphone, the researchers estimate the physical hardware bill of materials at about $400.

The smartphone is mounted directly above PhoneBot's hips, forming the robot's torso. The phone's camera handles visual tracking while onboard inertial sensors monitor movement and orientation. The handset executes movement-control software and transmits commands over Wi-Fi to a small onboard computer, which forwards instructions to 13 motors: six in each leg and one rotational torso joint that allows the camera to pan horizontally while keeping the robot's feet planted.

The UCLA team evaluated the setup across four Android handsets: a 2017 Honor 9, a Moto G 2024, a Moto G 2025, and a Samsung Galaxy A16 5G. All four models supported baseline sensing, vision, and balance controls, demonstrating that recent flagship hardware is not required. However, the Honor 9 could not run the application's onboard mapping feature because required Google services were unavailable on the device.

PhoneBot builds upon prior smartphone-robotics platforms such as OpenBot, a $50 wheeled system developed by Intel's Intelligence Systems Lab, extending the paradigm to balance-intensive bipedal walking. To prevent motor strain, the UCLA team trained the locomotion software in simulation with torque constraints. When initial models developed an asymmetric gait that loaded one leg unevenly, the researchers introduced mirrored left-to-right training examples, which accelerated learning stability and distributed motor loads evenly between both legs.

In physical trials, PhoneBot successfully walked across flat terrain, tracked people, and stood up from a face-down position. The current prototype lacks arms, and speech recognition and natural language processing still rely on an external laptop communicating through the phone's microphone and speaker. The researchers plan to make PhoneBot fully open source and explore on-device language models, lightweight arms, and uneven terrain navigation in future revisions.