Online path searching for autonomous robot navigation

"Blind" goal reaching, a common autonomous robot navigation task, is applied in highly dynamic and unknown environments. In this it differs from "heuristic" goal reaching, which makes use of a geometrical or topological environmental map. Traditionally, blind goal reaching combines both obstacle-avoidance (OA) and goal-seeking (GS) behaviors, yet this is not a sufficient way to obtain a smooth path. And even more seriously, if the robot meets a dead end, the "OA+GS" approach may cause the dead-cycle (or local minimum) problem. This paper proposes a novel approach, memory grid (MG), which imitates the human memory and decision making functions. MG-based online path searching (PS) behavior provides a novel alternative to blind goal reaching. The experiments, including tests on a real sonar-based robot navigating in dead ends, have demonstrated not only that the performance of the "OA+GS+PS" approach is superior to that of "OA+GS" navigation algorithms, but also that, unlike the traditional "OA+GS" approach, it can solve the dead-cycle problem.