Player Koi: A Robotic Chessboard That Watches, Thinks and Moves

A physical chess game requires a second player, and the commercial robotic boards that remove this requirement cost several hundred dollars and are closed systems. This paper presents Player Koi, a robotic chessboard built from commonly available laboratory parts. An overhead IMX219 camera on a Raspberry Pi 5 reads every square as empty, white or black with a small fine-tuned YOLOv8 classifier. The observed change is matched against every legal move, so the full position is tracked without recognising piece types. Stockfish picks a reply, and an Arduino Uno drives a CoreXY gantry beneath the board that drags the piece into place with an electromagnet. The camera then verifies that the move was executed. The system also removes captured pieces automatically and includes a move coach and a puzzle trainer. On held-out test images from three camera and lighting setups, the square classifier labels 99.5% of crops correctly (2,292 of 2,304). In ten full games, 312 of 330 human moves were read correctly, 16 were flagged for the player to fix and only 2 were resolved wrongly, while 318 of 326 robot moves were confirmed by the camera.