Your Old Phone Could Learn to Walk, Thanks to UCLA

The phone in your junk drawer could have a second career. Researchers at a Los Angeles university have built a small walking robot that treats a spare smartphone as its brain, camera, and sense of balance all at once — which is why the Old Phone Could Learn to Walk story is less about robots and more about what phones are actually capable of. The project, named PhoneBot, shows that a handset you would otherwise trade in or recycle can sense a room, watch a person move, and keep a two-legged machine upright without extra sensing hardware.

PhoneBot stands about nineteen inches tall and weighs roughly four pounds, including the phone strapped above its hips like a torso. The handset's camera watches the room and tracks people, its motion sensors read how the body tilts and turns, and the apps running on it issue the movement commands. Those commands travel over Wi-Fi to a small onboard board that passes them to the motors. All told, the robot has thirteen motors, six per leg plus one that rotates the torso so the camera can look sideways while the feet stay planted. In tests, the machine walked across a room, followed a person with its gaze, and stood itself back up from a face-down position.

The handset is the whole brain

Most robots split their thinking across purpose-built sensors, cameras, and processors that all have to be bought and wired together. PhoneBot collapses that stack into one device that already sits in billions of pockets. The research team tested four ordinary Android handsets, including one dating back to 2017, and found every one of them could handle the core sensing, vision, and movement-control jobs. The oldest phone in the group tripped on only one task, a mapping feature that needed services the handset no longer had. According to the team's paper, as reported by Tech Xplore, the hardware bill outside the phone comes to about four hundred dollars — remarkably low for a walking humanoid, and cheap enough that classrooms and student labs could actually build one.

That price point matters because humanoid robots are usually priced like lab equipment. Teaching a robot to walk means training its control software in simulation first, so the machine learns only movements its real motors can pull off. When the team trained PhoneBot this way, an odd habit appeared: the left foot kept drifting forward while the right leg carried most of the load, even though the robot was built symmetric. The fix was simple and a little elegant. Each training example was mirrored, swapping left and right, so the software learned from both versions at once. Runs with that mirror trick learned faster and spread the load evenly between the legs, and those controls carried straight over to the physical robot.

From desk clutter to lab equipment

This idea has a predecessor. A few years ago, engineers at an Intel research lab introduced OpenBot, a platform that bolted a smartphone onto a wheeled body for roughly fifty dollars. PhoneBot is the harder sequel: a machine that must balance on two legs instead of rolling on wheels. The open-source design is meant to be copied and adapted, giving schools and hobby labs a starting point instead of a research project that only giant labs can afford.

For now, the machine has limits worth stating plainly. PhoneBot currently walks on flat ground, carries little, and has no arms. Its small motors cap how much weight it can move and how hard it can push. Still, the trajectory is clear: the team wants to try uneven terrain and add lightweight arms, and spoken interaction is already partially supported through the phone's own microphone and speaker. That last feature still leans on an external laptop for understanding speech, so the next step is moving the language model onto the phone itself.

The wider point lands well beyond one research lab. Upgrade culture keeps handing us faster chips while the old ones still work, and reports of Gen Z e-waste awareness suggest younger buyers are already asking where those old devices go. A platform that turns yesterday's handset into today's lab instrument answers that question literally. The Old Phone Could Learn to Walk premise also pairs neatly with how people now shop for phones, given that a new budget phone wants to kill the flagship upgrade anyway: if the old phone still has a working brain, why retire it to a drawer?

There is a learning angle that should interest anyone paying attention to how AI spreads. The Old Phone Could Learn to Walk design treats the smartphone as a general-purpose sensor and computer rather than a gadget with one job. That is the same instinct behind personal AI agents that now speed up networked teams, where existing software gets asked to do far more than its original job. The difference is that this time the hardware is the hero: cameras, accelerometers, and processors that took a decade of phone wars to perfect are suddenly the cheapest robotics kit on the market. The research details were published in October via that outlet, and the paper itself is openly available for anyone who wants to build the thing.