On-line safe path planning in unknown environments

Abstract

For the on-line safe path planning of a mobile robot in unknown environments, the paper proposes a simple Hopfield Neural Network (HNN) planner. Without learning process, the HNN plans a safe path with consideration of "too close" or "too far". For obstacles of arbitrary shape, we prove that the HNN has no unexpected local attractive point and can find a steepest climbing path, if a feasible path(s) exists. To effectively simulate the HNN on sequential processor, we discuss algorithms with O(N) time complexity, and propose the constrained distance transformation-based Gauss-Seidel iteration method to solve the HNN. Simulations and experiments demonstrate the method has high real-time ability and adaptability to complex environments.