Interlaced Extended Kalman Filter for Real Time Navigation

Real-time applications ask for reduced computational cost algorithms. In robotic exploration of unstructured environments the problem is more challenging: several tasks, at the same time, must be carried on ranging from reactive behaviours to the building of a structured representation of the environment itself. Many sensor signals have to be processed at each step to estimate both landmarks and robot positions. This mapping aptitude can be implemented through an Extended Kalman Filter recently proposed in a previous paper. Due to the large number of estimated variables, and real-time constraints, the filter is better implemented in its interlaced version. The novelty of this paper consists in extending the IEKF filter, removing some hypothesis on the linearity of both state transition and observation mapping, in order to further reduce computational burden and then achieve a better tradeoff among computational load and accuracy.