RobLab: A testbed for integrated perception, planning, and control

We present the perception, planning, and control of RobLab, a robotic adaptation of the marble maze game 'Labyrinth.' We first present a framework which is capable of autonomously solving the maze by following a planned path. This is achieved using RGB threshold filters to detect the maze walls and track the marble, an A∗ search to generate a path, and a PID controller to guide the marble along the path. A Behavior Tree is then used to extend this framework to include additional behaviors which improve the system's robustness and enable the operator to override if requested. Experimental results and the adequacy of Behavior Trees as a tool for robotic task execution are discussed. This work aims to provide a case study in prototyping high-level robot behaviors which could later be deployed to a more complex system.