递归神经网络的进化机器人路径规划方法

Path planning based on a recurrent neural network for an evolutionary robot

针对机器人递归神经网络控制器在进化优化过程中存在的问题,利用改进的进化算法对递归神经网络控制器进行优化设计,提出了一种基于递归神经网络的进化机器人路径规划算法,该算法利用高斯变异和柯西变异相结合的方式进行变异操作,利用个体适应度和种群多样性指标使交叉概率和变异概率进行自适应调整.给出了算法的具体步骤,并与基于标准前馈网络的路径规划方法进行了比较.仿真结果表明递归神经网络控制器对动态未知环境具有更好的适应性.

To investigate path planning for mobile robots based on a recurrent neural network and evolutionary algorithms, a recurrent neural controller was trained via an improved evolutionary algorithm. An algorithm for path planning was developed based on a recurrent neural network for an evolutionary robot. A combination of Gaussian and Cauchy mutations was used to ensure larger mutation steps and escape from local minima. Crossover and mutation probabilities were adjusted automatically according to variations in the diversity of the population and the fitness of individuals. A detailed process to apply the algorithm was presented. The algorithm was compared with the standard feed-forward network-based method of path planning. Experimental results indicated that the recurrent neural controller has high adaptability to dynamic environments.