Truncated unscented particle filter for dealing with non-linear inequality constraints

This paper addresses state estimation where domain knowledge is represented by non-linear inequality constraints. To cope with non-Gaussian state distribution caused by the utilisation of domain knowledge, a truncated unscented particle filter method is proposed in this paper. Different from other particle filtering schemes, a truncated unscented Kalman filter is adopted as the importance function for sampling new particles in the proposed truncated unscented particle scheme. Consequently more effective particles are generated and a better state estimation result is then obtained. The advantages of the proposed truncated unscented particle filter algorithm over the state-of-the-art particle filters are demonstrated through Monte-Carlo simulations.