动态未知环境下一种Hopfield神经网络路径规划方法

Hopfield neural networks for path planning in dynamic and unknown environments

针对动态未知环境下移动机器人路径规划问题,采用一种有效的局部连接Hopfiled神经网络(HopfieldNeuralNetworks,HNN)来表示机器人的工作空间.机器人在HNN所形成的动态数值势场上进行爬山搜索法来形成避碰路径,并且不存在非期望的局部吸引点.HNN权值设计中考虑了路径安全性因素,通过在障碍物附件形成局部虚拟排斥力来形成安全路径.HNN的连接权是非对称的,并且考虑了信号传播时延.分析了HNN的稳定性,所给稳定性条件和时延无关.HNN模型中突出了最大传播激励,从而使得HNN具有更广的稳定性范围并能表示具有更多节点的机器人工作空间.为对该HNN有效仿真求解,结合约束距离变换和HNN的时延性,给出了单处理器上高效的串行模拟方案,规划路径的时间复杂度为O(N)(N是HNN中神经元的数目),使得路径重规划能快速在线进行.仿真和实验表明该方法的有效性.

To deal with the path planning of mobile robot in dynamic and unknown environment,an efficient and locally connected Hopfield neural network (HNN) is proposed to represent the workspace of the robot.The robot dynamically traced the numerical potential field of the HNN by hill climbing method to find the collision-free path without any unexpected local attractive points.The safety of the planned path was considered in the weight design of the HNN, and local virtual repulsive forces were formed around obstacles to generate safe path.The HNN model considered the time delay of signal diffusion and had asymmetric weights.The stability of the HNN was analyzed and the given stable condition of the HNN was independent on the time-delay of signal diffusion.Because the model emphasizes on the diffusion of maximal stimulation, the given stable condition is more relaxed and leads the HNN to represent a large workspace with more grids.To efficiently simulate the HNN,combining the constrained distance transformation and the delays in HNN,sequential simulation of the HNN on a single processor is proposed to plan path in \%O(N)\% time,where \%N\% is the number of the nodes of the HNN.The \%O(N)\% time complexity of sequential simulation accelerates the path re-planning on-line.The simulations and experiments demonstrate the effectiveness of the method.