Model-based sensor fusion and filtering for localization of a semi-autonomous robotic vehicle

Abstract

This paper refines a physically-inspired model governing the dynamic motion of a vehicle. We present a method used to perform experimental parameter calibration, and then use this model to build an observer (an extended Kalman filter). Experimental results with a robotic vehicle fitted with a prototype kit focus on recovering the truthful real-world information in the context of systematic errors (a faulty wheel encoder sensor), randomly occurring errors (a faulty ultrasonic sensor) and simplifying model assumptions (e.g. usage of two identical motors). We show that our model-based approach is able to perform reasonably well even under these extreme circumstances.