Flow Field and Neural Network Guided Steering Control for Rigid Autonomous Vehicles

This paper studies the steering control for low-speed manoeuvring of autonomous ground vehicles. A guidance method combining flow analogy and a neural network model is proposed to produce the proper angular velocity for the vehicle, which can be used as a reference for the control of the steering wheel. In a previous study, fluid flow itself has shown outstanding global search capabilities in guiding the vehicle through complicated environments. But the vehicle is not always able to follow the motion of the flow due to the difference of their nature. In this paper, the heat flow analogy is used instead of fluid flow, and a neural network model is added upon the flow layer in order to produce a steering reference more suitable for a rigid vehicle. Simulated results demonstrate that, except for the branching situations, the proposed method is able to guide the vehicle towards its desired destination.