AI Makes Its Move

Almost as long as there have been digital computers, engineers have been programming them to play chess. The quest to create a machine capable of mastering chess began in the mid-20th century. Early efforts were rudimentary, limited by the computational power available at the time. These pioneering programs could only evaluate a few moves ahead and were often easily defeated by even amateur human players.

The development of chess engines accelerated dramatically with the advent of more powerful hardware and sophisticated algorithms. By the 1980s, programs like Belle and Deep Thought were making headlines by winning against strong human players and in computer chess tournaments. The culmination of these efforts came in 1997 when IBM's Deep Blue defeated reigning World Chess Champion Garry Kasparov in a highly publicized match.

These chess-playing machines typically operate either entirely in a simulated environment, or they require human operators to enter the moves of the opponent, and also to move its own pieces on the board. That is not exactly like sitting down with a friend to play a nice game of chess, now is it? To make for a better experience, engineers at the Delft University of Technology have constructed a chess playing robot. It is capable of seeing and understanding the board, calculating moves, and physically moving the pieces. It is even equipped with generative AI capabilities so that you can chat during the game.

The system, called OpenChessRobot, is not just about fun and games. The researchers designed it primarily as an open-source tool that others can leverage in human-robot interaction studies.

On the hardware side, OpenChessRobot is equipped with an NVIDIA Jetson Nano computer for processing visual data, and a ZED2 StereoLabs camera for object recognition and depth perception. A Franka Emika Panda robot arm with a Franka Hand and a custom, 3D-printed gripper enable the system to physically interact in the game by moving pieces. A speaker and microphone are also included to support the robot’s conversational capabilities.

A number of software modules were developed to support this hardware. There is a perception module that recognizes the chess pieces and determines their locations in three-dimensional space. There is also a chess engine to analyze the present state of the game and determine the best move to make. When the best move has been found, a motion planning module directs the robot arm to carry it out. Finally, the interaction module, which leverages OpenAI’s ChatGPT, allows users to carry on some friendly banter with their robotic opponent during the game.

While this robot can play a mean game of chess, that was not the main point of the work. Looking ahead, the team plans to use OpenChessRobot to study how AI-embodied robots influence people during their interactions. All of the code and training data, as well as installation instructions and documentation are open source and available on GitHub. Aside from the robot arm, the components used in the system are fairly inexpensive. It was noted that other robot arms could be used, however, so there is hope for the hobbyist that wants to build their own chess-playing robot using similar techniques.