For years the trend has been towards automated and assisted driving, for which vehicle dynamic control is needed. Effective vehicle dynamic control requires exact knowledge of the vehicle motion state and the road conditions, which usually comes from state estimators. As vehicle dynamics models often are nonlinear, the design of those estimators can be a difficult task. The Unscented Kalman Filter is a powerful tool for nonlinear state estimation. Besides the selection of an appropriate model, the choice of covariance matrices is main task in the design phase. For this, optimization-based tuning has become a state of the art method. Often minimization of the estimation error is done, for example with global optimization methods [1]. However, this approach does not fully take into account the stochastic properties of the estimated variables, which can result in implausible covariance estimations. Addressing this problem, different approaches for optimization-based tuning arepresented in this paper using the example of an UKF state observer for a four-wheel drive electric vehicle [2], [3]. First, details on the used vehicle model as well as the estimator implementation are given. Then optimization and validation are carried out using data from driving tests in a vehicle simulation including various severe maneuvers and varying road conditions. Last, the results are compared and discussed.

Dr.-Ing. Hannes Heidfeld, Otto-von-Guericke-University Magdeburg, GERMANY Dr.-Ing. Martin Schünemann, Otto-von-Guericke-University Magdeburg, GERMANY