Robust and Precise Localization of Mobile Robots using Finite Impulse Response Estimation for Fusing Odometry with Position Measurements

The well known Kalman Filter (KF) and its nonlinear counterpart Extended Kalman Filter (EKF) are widely used for localizing mobile robots. Their main disadvantage is that their performance depends strongly on the probabilistic models of measurement and robot motion. An accurate probabilistic motion model is often unavailable, in this case the KF and EKF often demonstrate poor robustness and may diverge. The unbiased finite impulse response (UFIR) filter is an universal estimator for linear systems. The extended UFIR (EFIR) is the counterpart of the UFIR for nonlinear systems and operates similarly to the EKF. UFIR and EFIR utilize the most recent past measurements on a horizon of points and do not need any probabilistic model. In this paper a pose estimator for position measurements based on FIR algorithms is developed. The paper provides a comparative experimental analysis for robustness and accuracy of KF, EKF, UFIR and EFIR.