UKF/H-Infinity Filter for Low-Cost Localization in Self-driving Cars

Abstract

A self-driving car is one of the engineering marvels in recent time due to autonomous capabilities. An autonomous car can sense its surroundings through various integrated sensors and navigate the vehicle based on sensor data with Model Predictive Control System. Vehicle Navigation in self-driving car helps to determine the vehicle position by localization through high precision Global Navigation Satellite System and Inertial Navigation System (GNSS/INS) technologies with various sensors like Light Detection and Ranging (LIDAR), Radio Detection and Ranging (RADAR) and Global Positioning System (GPS) Sensor. Vehicle Navigation is completely depending up on the localization of the vehicle. The most popular method like particle filter and Kalman filter-based approach lacks stability with random particle distribution, and some of the GNSS-based methods are very expensive in terms of implementation. So here introducing an unscented Kalman filter and H-Infinity filter (UKF/H-Infinity)-based novel approach for the low-cost localization technique in self-driving cars. This robust control model helps to handle the position error from the GPS sensor with H-Infinity filter and UKF by the help of Constant Turn Rate Acceleration (CTRA) motion model. This work includes the detailed discussion of pose estimation of vehicle with better yaw stability and minimum computational overhead in estimation.