Simulated control of a tracking mobile robot by four aVLSI integrate-and-fire neurons paired into maps

Abstract

A simulated four-wheeled robot is controlled exclusively by four aVLSI integrate-and-fire neurons paired into winner-takes-all maps. The neural network takes analog sensor data as input and outputs to stepper motors controlling steering and throttle. The robot follows a randomly moving target in a closed environment 67% better than by chance, based on average distance to target. Simulation results suggest that silicon neural networks based on biological computing principles are effective, efficient, and compact embedded controllers. Test results should be confirmed on a physical implementation of the robot, and research should continue in network and circuit optimisation, as well as in the creation of robot societies.