Localization Confidence Domains via Set Inversion on Short-Term Trajectory

The knowledge of localization uncertainties is of prime importance when the navigation of intelligent vehicles has to deal with safety issues. This paper presents a robust estimation method that is able to quantify the localization confidence based on interval analysis and constraint propagation. First, tightly coupled position domains are computed by constraint propagation on Global Positioning System (GPS) measurements and a precise 3-D map of the drivable area. Since GPS is prone to satellite masking and wrong measurements in urban areas, a second stage provides localization integrity and information availability by the use of a position and proprioceptive data history. A robust constraint propagation algorithm is employed to compute the current vehicle pose. It is able to handle erroneous positions with a chosen integrity risk. Experiments carried out in urban canyons illustrate the performance of the method in comparison with a particle filter. Despite bad satellite visibility, full positioning availability is obtained, and errors are less than 5.1 m during 95% of the trial. In opposition to the particle filter, confidence domains are consistent with ground truth, which confirms the high integrity of the method.