The Tekkotsu robotic software framework's perception and manipulation primitives have largely been confined to planar worlds. In this thesis I bring Tekkotsu out of the plane by solving a three-dimensional manipulation task: autonomously playing chess on a real board. The project used a new version of the Chiara hexapod robot with a gripper customized for tournament-standard chess pieces. I developed techniques for detecting pieces, for using multiple images to deal with occlusions, for inferring opponent moves from noisy information about changes in square occupancy, and for localizing the robot with respect to the board. Accurate localization of pieces and the robot itself were achieved by a camera alignment procedure that produced a homography correction matrix that was then applied to the robot's camera projections. The Chiara's limited reach required moving its body relative to the board. I developed motion strategies for positioning the robot close to the board while keeping the legs from intruding into the playing area. Pieces were modeled as vertical cylinders of varying heights, and manipulation planning algorithms were developed to execute moves, including captures, using an optimal combination of arm trajectories and body motions. The robot successfully competed in the AAAI-2010 Small-Scale Manipulation Challenge. As a result of this work, Tekkotsu's dual-coding vision system can locate objects more accurately on its world map, and its manipulation planner has become more sophisticated. The techniques developed here can be applied to similar manipulation tasks, such as playing other board games.

49 pages

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by [email protected]