Wheeled Mobile Robot Modeling for Local Navigation Using System Identification

Abstract

In this work we investigate wheeled mobile robot (WMR) kinodynamic modeling using system identification techniques. The larger context is the development of a novel local navigation algorithm based on long kinodynamic trajectories which account for vehicle drivability. A high-fidelity physics-based differential-drive robot model is used to generate command velocity and ground truth data sets, which are subsequently analyzed using multiple system identification models. Parametric linear ARX and ARMAX models are considered. Experimental results using multiple training and validation data sets corre-sponding to different driving scenarios and vehicle loading are explored. The resulting model-generated WMR paths demonstrate excellent performance.