Konstantinos Tsianos, Real-time motion planning under realistic kinodynamic constraints

Real robots have kinodynamic constraints in their motion, imposed by laws of physics and motor limitations. Planning for such robots can become challenging especially when the robot is only dynamically stable (e.g. segway). Moreover, in realistic scenarios, mobile robots need to operate in environments that are partially known, or contain moving obstacles. In this work a motion planning algorithm that deals with all those issues is presented. The algorithm samples random controls and uses forward integration in a physics simulator, to produce collision-free feasible motions. Furthermore, the planner operates in real-time, and revises the plan executed by the robot at regular time intervals. A subdivision scheme is used to efficiently explore the part of the space within the robots sensing range, and a low dimensional adaptive potential field is used to provide a general sense of direction towards the goal.Our experiments exhibit how an autonomous segway can navigate safely in static environments, going through narrow passages, without ever falling. Experiments in progress provide promising evidence that the same algorithm can be used in the presence of moving obstacles in the workspace.