基于自组织LMBPNN的移动机器人路径规划器

Obstacles avoidance path planning method based SOM-LMBPNN for mobile robot

提出一种自组织LMBP神经网络,并将之用于移动机器人免碰路径规划。该算法首先用基于距离传感器的底层局部路径规划器生成初始路径,然后用自组织神经网络将该路径进行样本数据分类,之后将自组织神经网络的权值作为LMBP的输出样本,移动机器人的起始点与目标点作为LMBP神经网络的输入样本进行学习。这样,不但解决了三层LMBP样本若庞大则增加存贮、运行成本,以及数据冗余问题,并且随着机器人对未知环境探索的增多,所构建的地图越趋丰满。仿真结果说明该方法很好效。

This paper presents a path-planning algorithm based on a self-organizing feature map-Levenberg-Marquardt Back propagation Neural Network（SOFM-LMBPNN） for a mobile robot with static obstacles environments.The algorithm generates a original path using a base path planner based on range sensors firstly,then classifies the path using a one-dimensional self-organizing feature map neural network,lastly trains the LMBPNN with the start configuration and goal configuration as the input samples and the weights of the SOMFNN as the output samples.This algorithm not only reduces the cost of the store and operation,but also solves the problem of the redundancy samples to some degree.The simulation experiments verify the efficiency of this algorithm.