A neural network approach for solving the path planning problem

A neural network technique for robot path planning on a grid model of terrain has been developed. Given a start grid and a goal grid, a direction map is constructed by using the information of the grid map. Each grid is associated with an individual element in the direction map which represents the best direction of travel with respect to the goal grid. Using the direction map, the path from the start grid to the goal grid can be determinated by following the direction indicated on the direction map. The algorithm for computation of the direction map can be implemented in a modified Hopfield network where each M-P neuron is replaced by a mesh processors. The mesh processor is constructed by using logic gates including AND gates, an OR gate, a NOT gate and flip flops. Simulation results of path planning using the proposed technique are provided