Vehicle Parameter Estimation with Kalman Filter Disturbance Observer

Abstract

Vehicle parameter estimation is an identification problem which is almost exclusively solved with least squares methods. In practical applications, some prerequisites for ordinary least squares are not satisfied, such as the assumption of an error-free regressor and independent and normally distributed noise. In this work, three parameter estimators for dynamic systems are presented which address these problems: a total least squares estimator to deal with regressor errors, and a robust M-type Kalman filter for long-tailed error distributions from the literature are implemented. As the main contribution, a disturbance observer parameter estimator (DOPE) is proposed which observes a disturbance process instead of assuming independent noise in addition to the parameter estimation. The estimators are applied to a mass and friction identification problem for an electric tractor on a large number of experiments. The DOPE has the lowest median estimation error and the lowest sensitivity to bad measurement data. The results show that the disturbance observer is a valuable tool for practical Kalman filter parameter estimation.