Sensor Fusion Algorithm Selection for an Autonomous Wheelchair Based on EKF/UKF Comparison

Abstract

—This paper compares two sensorial fusion algorithms based on their characteristics and performance when applied to a localization system for an autonomous wheelchair in dynamic environments. The mobile robot localization module is composed by three sensors: Encoders attached to the wheels, LIDAR and IMU. The information provided by each one is combined according to their covariance obtaining the most reliable pose estimation possible. For this purpose, it focuses on the study of two fusion algorithms, the Extended and Unscented Kalman filters, detailing their properties and operation. Both methods are implemented in the wheelchair for its comparison. The experiments carried out demonstrate how the localization results with UKF are more precise than using the EKF in a non-linear system and shows similar pose estimation when using a constant velocity model, despite the fact that the UKF needs longer execution time than the EKF.