Mobile vehicle navigation in unknown environments: a multiple hypothesis approach

The paper describes an algorithm for sensor-based map building and navigation for an autonomous mobile vehicle. The algorithm is based on the use of an extended Kalman filter to obtain estimates of the location and identity of geometric features in an unknown environment. A multitarget tracking methodology is applied to the evaluation of multiple hypotheses about the locations of geometric features in the environment. The algorithm does not require any a priori information about the environment. It is capable of initiating new geometric features and identifying the type of a geometric feature from the given set of geometric features, utilising the data provided by a set of sonar sensors. The algorithm is also capable of deleting geometric features from the map of the environment when they are no longer detected by the sensors. The implementation of the algorithm is discussed, and results using real sonar data are presented.