ADMSV - A Differential Machine Learning based Steering Controller for Smart Vehicles

Abstract

Electric power-assisted steering (EPAS) is a mechanism of using electric power to enhance the efficiency, performance, and reliability of steering operations in vehicles. In the modern-day fully-autonomous and semi-autonomous vehicles, the real-time operation of EPAS systems has challenges related to the unmodeled dynamics, irregularity of the system operation, and variable road conditions. In this paper, a machine learning-based control system (ADMSV) that incorporates motion-related inputs such as direction, velocity, and torque has been developed to optimize and improve the overall efficiency of electric power-assisted steering in intelligent vehicles. The proposed system is used to calculate numerous external inputs and generate steering-related outputs (angular velocity, angular difference, output torque) which could help supply the adequate amount of torque that helps the vehicle to maneuver the wheels more easily or comfortably depending on various road and driving conditions.