Quantifying the Risk of Unmapped Associations for Mobile Robot Localization Safety

Abstract

Integrity risk is a measure of localization safety that accounts for the presence of undetected sensor faults. The metric has been used for decades in aviation and has recently been applied to terrestrial robots operating in life-critical missions. For ground vehicles, integrity risk can be quantified for systems using lidar measurements, where two specific fault types have been identified: miss-association and unmapped association. While miss-association faults, which occur when a correctly extracted feature is associated to the wrong landmark, have been well-studied, the probability of an unmapped association fault, where an incorrectly extracted feature is associated to a landmark, is not well-understood. Namely, previous research has never quantified this value and instead relies on an assumed value, one whose value has not been properly justified. This work is the first to provide a methodology that estimates the risk of unmapped association for each mapped landmark; the paper demonstrates the effect of this probability for both the chi-squared and fixed-lag smoothing methods for integrity monitoring. Data collected in downtown Chicago, IL USA was used to test the impact of unmapped association faults on localization safety. The results indicate that using the previously assumed value is reasonable in many situations, but that applications with strict safety requirements should incorporate the method described here to properly account for unmapped association faults.