Monte Carlo Localization with Hilbert maps as Likelihood Fields

Sammanfattning: This work focuses on using sparse Hilbert maps as continuous likelihood field observation models, instead of utilizing traditional beam observation models. This observation model is integrated with a Monte Carlo Localization scheme, and localization is simulated in randomly generated 2D mazes. When compared with the beam observation model, this change of observation model shows a 30% improvement in the positional accuracy of tracking as well as a 25 times decrease in the run time at the cost of a reduced success rate for global localization. However, using the likelihood field model for Hilbert maps experimental results do not demonstrate any particularly strong robustness against sensor inaccuracy, a prevalent issue in cheaper or more lightweight sensors.