Implementation of a neuron-based autonomous Mobil Robot on a DSP

Abstract

An autonomous mobil robot has been implemented on a Digital Signal Processor (for real time operation) using neural networks as the main part of the program that runs on the processor. The neural network was based on a single layer perceptron (SLP) with two neurons, four inputs coming from four sonar sensors, and two outputs to control the direction of two CD motors. The goal of the mobil robot is to avoid obstacles while it runs randomly in a given room. Two training sets were tested to provide two di®erent reactive behaviors, the first one with forward direction, left turn and right turn, and the second which includes reverse direction. Both behaviors were compared.