A traction Control System based on Co-evolutionary Learning in Spiking Neural Networks (SNN)

Abstract

A traction control system is designed and trained for different road conditions with co-evolutionary learning based on a genetic algorithm. Common solutions do not consider the variation and oscillation created in the transition between roads defining a control logic which is highly dependent on road accuracy and a speed estimator. To solve this problem, a co-evolutionary learning process is used. This procedure trains the control algorithm, a spiking neural network, on different roads and transitions looking for the worst-case scenario. We have developed a control algorithm with a good dynamic response to constant and changing roads. This control algorithm makes the system stable when the road estimation is delayed or unstable, solving a common flaw produced by sensor noise or computation delays.