Mobile Robot Self-localization Based on Multi-sensor Fusion Using Limited Memory Kalman Filter with Exponential Fading Factor

Abstract

Accumulative errors can be retained all the time when classical Kalman filtering is adopted for odometer-based dead reckoning, thereby affecting self-localization accuracy of the robot. A mobile robot self-localization method based on limited memory Kalman filtering (LMKF) with exponential fading factor was proposed to reduce accumulative errors of the odometer and improve localization accuracy of the mobile robot. The self-localization system of mobile robot was built. A mathematical model was established based on LMKF with exponential fading factor. A dead reckoning method fusing multi-sensor information was proposed. The model accuracy was verified through simulation and test. Results indicate that the LMKF method with exponential fading factor positively affects the tracking of high-speed maneuvering dynamic targets, and its localization error is reduced by 42.5% compared with the odometer-based dead reckoning. The tracking accuracy of the mobile robot is stable at 0.5 m. This study can provide references for mobile robot selflocalization using multi-sensor.