Application of the Kalman-particle kernel filter to the updated inertial navigation system

This paper considers a new nonlinear filter which combines the good properties of the Kalman filter and the particle filter. Compared with other particle filters like Rao-Blackwellised particle filter (RBPF), it adds a local linearization in a kernel representation of the conditional density, which yields a Kalman type correction complementing the usual particle correction. Therefore, it can operate with much less number of particles. It reduces the Monte-Carlo fluctuations and the risk of divergence. The new filter is applied to the highly nonlinear and multimodal terrain navigation problem. Simulations show that it outperforms the RBPF.