PID Control for a Path-Following Error-Producing Neural Network

Abstract

In this paper, a neural network-based methodology is proposed with the goal of enabling a mobile robot to guide itself along a distinct path. By feeding a stream of pre-processed path images through a neural network, an error can be produced for a PID controller to direct a three-wheeled differential-drive robot along a path. An MLP network trained with a direction-classification dataset and a classical PID controller are utilized. Practical trials confirm the effectiveness of the PID stabilization on constant direction paths.