Autonomus Systems Control Design Using Neuro-Evolution

This paper aims to design and implement a neuro-evolution for the purpose of an autonomous vehicle moving on the road, where its control is performed by an MLP-type feed-forward artificial neural network. The network perceives the environment and controls the angle of the vehicle's wheels as well as its speed in a 2D simulation environment. The vehicle itself consists of a four-wheeled chassis with rotatable front wheels with its own dynamics. The neural network is represented by a four-layer architecture and its parameters are optimized using a parallel genetic algorithm to minimize the chosen performance criterion for controlling the motion of the vehicle.