Localization accuracy improvement of autonomous vehicles using sensor fusion and extended Kalman filter

Abstract

Advanced driver assistance systems and autonomous vehicles rely heavily on position information, therefore, enhancing localization algorithms is an actively researched field. Novel algorithms fuse the signals of common vehicle sensors, the inertial measurement unit and global positioning system. This paper presents a localization algorithm for vehicle position estimation that integrates vehicle sensors (steering angle encoder, wheel speed sensors and a yaw-rate sensor) and GPS signals. The estimation algorithm uses an extended Kalman filter designed for a simplified version of the single track model. The vehicle dynamics-based model only includes calculation of the lateral force and planar motion of the vehicle resulting in the minimal state-space model and filter algorithm. A TESIS veDYNA vehicle dynamics and MathWorks Simulink-based simulation environment was used in the development and validation process. The presented results include different low- and high-speed maneuvers as well as filter estimates of the position and heading of the vehicle.