Maximum likelihood estimation-assisted ASVSF through state covariance-based 2D SLAM algorithm

Abstract

The smooth variable structure filter (ASVSF) has been relatively considered as a new robust predictor-corrector method for estimating the state. In order to effectively utilize it, an SVSF requires the accurate system model, and exact prior knowledge includes both the process and measurement noise statistic. Unfortunately, the system model is always inaccurate because of some considerations avoided at the beginning. Moreover, the small addictive noises are partially known or even unknown. Of course, this limitation can degrade the performance of SVSF or also lead to divergence condition. For this reason, it is proposed through this paper an adaptive smooth variable structure filter (ASVSF) by conditioning the probability density function of a measurement
to the unknown parameters at one iteration. This proposed method is assumed to accomplish the localization and direct point-based observation task of a wheeled mobile robot, TurtleBot2. Finally, by realistically simulating it and comparing to a conventional method, the proposed method has been showing a better accuracy and stability in term of root mean square error (RMSE) of the estimated map coordinate (EMC) and estimated path coordinate (EPC).