未知环境下移动机器人安全路径规划的一种神经网络方法

针对未知环境下移动机器人的安全路径规划,采用了一种局部连接Hopfield神经网 络(Hopfield Neural Networks,HNN)规划器;分析了HNN稳定性,并给出了存在可行路 径的条件.如果存在可行路径,该方法不存在非期望的局部吸引点,并在连接权设计中兼顾 "过近"和"过远"来形成安全路径.为在单处理器上有效地在线路径规划,采用多顺序的 Gauss-Seidel迭代方法来加速HNN势场的传播.结果表明该方法具有较高的实时性和环境 适应性.

A Neural Networks-based Approach to Safe Path Planning of Mobile Robot in Unknown Environment

For safe path planning of mobile robot in unknown environment, the paper uses a local connected Hopfield neural network (HNN) planner. The stability of the HNN is analyzed, and the condition for the existence of the feasible path(s) is given. If a feasible path(s) exists, the HNN does not have any unexpected local attractive point. The connected weight design considers both "too close" and "too far" to plan the safe path. For the HNN to on-line plan a path on a sequential processor, multi-sequential Gauss-Seidel iteration is used to accelerate the propagation of the HNN potential field. Results demonstrate that the method has good real-time ability and adaptability to environments.