Abstract
This paper considers a method for estimating vehicle handling dynamic states in real-time, using a combined system identifier / observer which attempts to maximise the information available from an incomplete system model and reduced sensor set. The study is carried out in simulation using two models; the 'source' model - used to provide regression data for identification and to validate the resulting estimator - considers yaw, sideslip and roll degrees of freedom and incorporates physically realistic nonlinear lateral tyre characteristics. The Kalman filter is based on a simple linear yaw/sideslip model, with two sensor configurations, both using lateral accelerometers only. It is assumed that reasonable estimates exist for the linear model inertias and geometry, with cornering stiffnesses adapted in an on-line identification process executed in conjunction with the Kalman filter. The adaptive observer provides acceptable performance, particularly in reconstructing yaw rate without a rate sensor. Some problems exist however; the filter induces low frequency errors that will require further analysis to eliminate.