Error-Tolerant Switched Robust Extended Kalman Filter With Application to Parameter Estimation of Wheel-Soil Interaction

Abstract

A real-time soil parameter estimation method for wheel motion control of a planetary rover is proposed. In this method, dominant soil parameters are updated in real time to compensate the error of the other parameters that are fixed by empirical typical values. Sinkage exponent and internal friction angle are the dominant characteristic parameters of wheel-soil interaction, because they dominate the normal stress and sheer stress that constitute the interaction forces on the contact surface. To estimate these two parameters in real time, a robust extended Kalman filter with error-tolerant switch (ETS-REKF) is proposed in this paper. The developed ETS switches filtering mode between robust and optimal, which is triggered when the magnitude of the sum of system errors reaches a certain threshold that is controlled by an error-tolerant factor (ETF). The proposed filtering algorithm can follow soil-type change quickly and in the meantime provide smooth parameter estimation without being sensitive to the inherent error of the empirical wheel-soil interaction model. From a sufficient condition of the filter stability, the range of the ETF is derived. In addition, a trigonometric approximation method is provided to simplify the original wheel-soil model for convenient utilization of the filtering algorithm. Simulation and experiment results have demonstrated that the proposed ETS-REKF has the advantages of both optimality of the EKF and robustness of the usual robust EKF, working efficiently in the presence of soil-type change and inaccuracy of the empirical wheel-soil model.Index Terms-Error-tolerant switch (ETS), parameter sensitivity, real-time estimation, robust extended Kalman filter (REKF), wheel-soil interaction.

Recent developments in terrain identification, classification, parameter estimation for the navigation of autonomous robots

“…These parameters are preferred over the slip ratio as the latter is ineffective in rough terrains. This work provides an improved algorithm of the work by Yuankai Li [57]. The real-time terrain estimation by two-layer process improves the performance of Extended Kalman filter [58,59] and Recursive Gaussian Newton algorithm [60].…”