Deep Learning-based Heading Angle Estimation for Enhanced Autonomous Vehicle Backward Driving Control

This paper presents an enhanced autonomous backward driving control system for a commercial vehicle. Autonomous backward driving involves accurately retracing a previously driven forward path of approximately 54 meters. Accurate estimation of the vehicle’s heading angle significantly impacts the control performance; therefore, this study proposes a sensor fusion approach utilizing embedded sensors available on commercial vehicles. A deep neural network-based intelligent technique dynamically assigns optimal weight values to sensor outputs, effectively improving heading angle estimation accuracy across diverse driving paths. The proposed method significantly reduces deviation errors between forward and backward trajectories to less than 0.5 meters by intelligently learning path-specific weight values through neural network training. Experimental evaluations conducted on various road conditions confirm the effectiveness and feasibility of the proposed deep learning-based autonomous backward driving control system.