Efficient geometrical obstacle reshaping for inevitable collision states avoidance

Classical motion planners and most of navigation routines discard dynamics of the system and therefore fail in some special conditions. If a robot is moving towards an obstacle with a high acceleration and speed, collision is inevitable because the obstacles are taken into account only either once they are in range, or a physical collision is reported by detection routines. In this paper, we have introduced a new approach named T-Bone algorithm, that considers kinematics of an extended case of a Dubins car moving in a static environment, in order to avoid potential collisions, however it inherently can be extended for more complex systems. This method deforms the perception of the collision sensors - which traditionally discard dynamics of the system and only consider physical appearance of the obstacles. Deformed state space is then fed into the planner and navigation unit in order to move the vehicle, safely. Bounded complexity of the algorithm makes it proper to be used in realtime applications. Implementation results of the algorithm and some typical test runs are presented in the paper.