Multi sensor fusion and constraint propagation to localize a mobile robot

Abstract

In this paper, a mobile robot localization method is presented like a constraint propagation problem. The mobile robot is equipped with an exteroceptive sensor and dead-reckoning and knows a map of its evolution environment. To model imprecision measurements of these sensors, data are given under shape of intervals. Our localization strategy is based on multi-target tracking, the tracks representing primitives of the robot's environment. During the robot's displacement, the algorithm tries to match its observations with the managed tracks or tries to initialize news with observations thanks to the theoretical map. The matching criterion used is founded on an overlapping percentage between corresponding subpavings. These two actions provide us some mathematical relations between the Cartesian coordinates and the associated measurements in comparison with the robot position (each data being modeled by intervals to manage naturally the imprecision). These data are fused by constraint propagation on intervals. This way composed of two steps forward and backward propagation allows to delete inconsistent values in intervals in order to reduce the imprecision. So, at the end of the localization process, we get a 3-D subpaving, which is supposed to contain the robot's position in a guaranteed way.