Guaranteed dynamic localization using constraints satisfaction techniques reaching global consistency

This article deals with data fusion using ensemblist tools in general, and constraints satisfaction techniques on real intervals, in particular. Indeed, such an approach seems to be well adapted if the data presents a strong redundancy, if the equations are non linear and if the real time implementation on a computer is a key issue. The contribution of this work is primarily methodological as we propose an original method to reach global consistency in a calculable number of arithmetic operations. From the application point of view, we are interested here in the state estimation process of the kinematics state of a car using the measurements of the four ABS sensors, the angle of the driving wheel and a GPS receiver. Experimental results illustrate the performance of such an approach in comparison with the extended Kalman filter.