Vehicle localisation in a poorly known environment

The aim of this paper is to present an algorithm for the localisation of a vehicle in an unknown or partially known environment. The basic localisation algorithm uses extended Kalman filtering to fuse internal movement information and observations of absolute references. These references are artificial beacons or specific environment elements with known positions. In this paper we propose an extension of this localisation algorithm which allows taking into account beacons with poorly known positions and to dynamically calibrate their positions. This calibration is also carried out by Kalman filtering. We introduce a criterion to decide if a beacon observation is used to estimate the position of the vehicle or the position of the beacon. We also propose a method to select the most significant observations. The algorithm has been studied for the example of a mobile indoor platform, but it can be applied as well to the navigation of underwater vehicles. The proposed method is an important step towards combining localisation and dynamic environment modelling.<
>