Guaranteed Robust Nonlinear State Estimator with Application to Global Vehicle Tracking

This paper deals with guaranteed recursive state estimation in a bounded-error context with application to global dynamical vehicle tracking. As in Kalman or approximate Bayesian filtering, prediction and correction phases alternate. A distinctive feature of the method advocated here is that its results are guaranteed, in the sense that the statements made about the possible values of the state vector are mathmatically proved, although all calculations are performed approximately on a computer. Sets will thus be provided that are guaranteed to contain all values of the state that are consistent with the information available and the bounds assumed on the state perturbations and measurement errors. Complexity issues are addressed and some tools are provided to facilitate real-time implementation. Results obtained with an actual vehicle are reported.