Combined state and parameter estimation of vehicle handling dynamics

This paper considers an extended form of the wellknown Kalman filter observer, to reconstruct dynamic states from a small sensor set, but also to rapidly adapt selected parameters in the nonlinear dynamic model which lies at the heart of the observer. A generic procedure is described for constructing the extended Kalman filter in such a way that any combination of model parameters can be identified. The study is carried out in simulation, using two different vehicle dynamic models, one to act as the test vehicle, the other forming the nucleus of the observer. The assumption is that while in-vehicle testing is most desirable for proving many controller algorithms, here we need ‘true’ reference state information, to examine Kalman filter accuracy. A number of experiments are carried out to prove the system’s identification properties and also to compare its performance with a more conventional Kalman filter, based on a linear handling model. The results demonstrate high levels of performance and significant robustness to design parameters such as parameter adaptation speed and anticipated sensor noise. Most significantly, the observer also operates well and is capable of parameter adaptation when model and sensor covariance information is not available – usually a restricting factor in practical Kalman filter estimator design. The only significant caveat is that we are ‘buying’ excellent dynamic tracking from a small sensor set, at some computational expense.