The application of Monte Carlo techniques to Bayesian state estimation is d
iscussed. A simple theory for the Monte Carlo uncertainty is developed show
ing that the number of Monte Carlo replications does not in principle have
to be large. A recursive on-line algorithm based on rejection sampling is g
iven and improved versions suggested. The methods are illustrated on a non-
linear pendulum with measurement saturation. (C) 2000 Elsevier Science Ltd.
All rights reserved.