This master's thesis addresses the development of a collaborative robotic application for playing chess, which enables the user to play a physical game against a robotic manipulator. Since modern chess engines play optimally and do not imitate humans, in addition to a classical chess engine we also developed a personalized artificial intelligence model that imitates the playing style of a selected grandmaster.
We designed and built a robotic cell that comprises the collaborative robot UFactory xArm6 with a Gimatic electric gripper and soft fingers made according to the fin-ray principle, a DGT electronic chessboard and chess clock, the DragonEYE rotary knob, and a Raspberry Pi 5 microcomputer as the central coordinator. We designed the software as multithreaded. The system recognizes the user's moves by comparing the state of the physical board with all legal moves from the previous position. The user can play against the classical chess engine Stockfish or against the personalized model VidmarAI, which we developed from the Maia Chess project through transfer learning on the games of grandmaster Milan Vidmar and which runs on the Leela Chess Zero engine. We evaluated the cell's safety according to the ISO 10218-2:2025 standard for three collision scenarios between the robot and the user.
On a held-out set of Vidmar's games, the VidmarAI model achieved a higher move-prediction accuracy than the base Maia-1900 model, and cross-evaluation confirmed that the improvement is specific to Vidmar's style. The safety measurements showed that soft fingers with a larger contact area keep forces and pressures within the limits specified by the standard. The developed application is a comprehensive example of collaborative robotics that integrates mechanical, electronic, software, and artificial intelligence components into an interactive system.
|
