HandBot is a robotic hand built by Deeplocal in collaboration with Google's Android Things platform, and it does exactly one thing: plays Rock Paper Scissors against you. You hold your hand in front of its camera, it recognizes your throw, decides its counter, and responds with servo-driven fingers. The whole cycle takes less than a second.
The hardware is a DIY kit. You assemble the mechanical hand yourself — frame, servos, linkages, wrist — then mount the controller and camera, load the software image, and calibrate the finger positions until the Rock, Paper, and Scissors shapes are clean and distinct. Google put together build documentation for it. The estimated build time for a first-timer is about seven hours. The parts cost runs around $490 for the hand and servo hardware, plus the Android Things dev board on top of that.
The vision system is what makes it work. A front-mounted camera feeds real-time image recognition to classify your throw. The system identifies your gesture, runs the game logic, and fires the servo commands — all on-board, no external server. Independent control over all five fingers means the output shapes are unambiguous. When it throws Paper, you see five extended fingers. When it throws Scissors, two fingers. The articulation is cleaner than what most casual human players produce.
For serious RPS training, HandBot solves a specific problem: getting enough throws in. Pattern conditioning and timing calibration require repetition, and finding a human partner willing to throw a thousand times in a practice session is not always practical. HandBot is available any time and doesn't get tired or bored or start throwing random shapes because its attention drifted.
The machine learning component is worth noting. HandBot tracks your previous throws and updates its behavior based on what it observes. This is exactly the kind of opponent modeling that competitive RPS players do manually — tracking your opponent's sequences and adjusting your throw selection based on their patterns. Against HandBot, you're training against a system that is actively doing the same thing to you.
For WRPSA events, HandBot operates as a practice and demo tool, not a sanctioned opponent. Official matches are human versus human, standard cadence, legal throws. But as a training environment and a demonstration of how machine vision applies to hand gesture classification, it's one of the more polished implementations anyone has built.