UCLA researchers build $400 open source humanoid robot powered by old Android phones

Researchers at the University of California, Los Angeles, have introduced PhoneBot, an open-source humanoid robot that bypasses the need for expensive onboard computers and dedicated cameras by utilizing old Android smartphones. Standing at 483 mm tall and weighing 1.8 kilograms, the platform costs approximately $400 in robotic hardware. The system relies entirely on the smartphone's built-in sensors and optics to handle core computing and vision tasks. Four different Android devices, ranging from a 2017 Honor 9 to budget models priced between $50 and $80, were successfully tested to run the application's core functions. These include visual person detection, pose estimation, motion sensing, and walking control via TensorFlow Lite.

Instead of rotating its entire frame, PhoneBot uses a dedicated joint to rotate the torso and the mounted smartphone, allowing the camera to track people dynamically. A separate internal computer communicates with the phone over Wi-Fi, translating wireless commands into joint movements across 13 servo motors. Six actuated joints per leg govern the hips, knees, and ankles, while an additional joint handles the torso rotation. This configuration enables the robot to execute basic walking maneuvers, recover from a face-down position, and interact visually with individuals without heavy mechanical overhead.

The research team trained PhoneBot's walking controller using reinforcement learning in a simulated environment, rewarding stability while penalizing erratic posture and excessive motor strain. Early physical trials revealed uneven gait patterns, which the team resolved by introducing mirrored training samples to enforce left and right movement symmetry. Across five independent evaluations, this method significantly reduced joint-angle symmetry error and torque imbalance. Despite these advancements, the platform still faces constraints from low-powered servos that limit its payload capacity, lacks manipulative arms, and is currently limited to flat-ground mobility.

The complete platform design is openly available for makers and researchers looking to replicate, modify, or investigate smartphone-driven robotics. By repurposing obsolete consumer electronics, the project offers a low-cost entry point for experimental robotics research.