Key points are not available for this paper at this time.
The authors describe an approach to mobile robot navigation that unifies the problems of obstacle avoidance, position estimation, and map building in a common multi-target tracking framework. Model-based navigation is viewed as a process of tracking naturally occurring geometric targets or beacons. Targets that have been predicted (expected) from the environment map are tracked to provide vehicle position estimates (localization). Targets that are observed, but not predicted, represent unknown environment features or obstacles and cause new tracks to be initiated, classified, and ultimately integrated into the map. A good sensor model is a crucial component of this approach, and is used both for predicting expected observations and classifying unexpected observations. This navigation framework is being implemented on a mobile robot that employs sonar as the principal navigation sensor. An implementation of model-based localization that achieves robust position estimation in a known environment is presented. Preliminary results in obstacle identification and map building are given that lead one to believe that a complete navigation system, encompassing localization, obstacle avoidance, and map building, can be implemented exclusively with sonar.>
Leonard et al. (Mon,) studied this question.