An extended Kalman filter for localisation in occupancy grid maps

The main contribution of this paper is an extended Kalman filter (EKF) based framework for mobile robot localisation in occupancy grid maps (OGMs), when the initial location is approximately known. We propose that the observation equation be formulated using the unsigned distance transform based Chamfer Distance (CD) that corresponds to a laser scan placed within the OGM, as a constraint. This formulation provides an alternative to the ray-casting model, which generally limited localisation in OGMs to Particle Filter (PF) based frameworks that can efficiently deal with observation models that are not analytic. Usage of an EKF is attractive due to its computational efficiency, especially as it can be applied to modern day field robots with limited on-board computing power. Furthermore, well-developed tools for dealing with potential outliers in the observations or changes to the motion model, exists in the EKF framework. The effectiveness of the proposed algorithm is demonstrated using a number of simulation and real life examples, including one in a dynamic environment populated with people.