Paper: Map-Aware Particle Filter for Localization

Paper ICRA2018: Map-Aware Particle Filter for Localization

This work presents a method to improve vehicle localization by using the information from a prior occupancy grid to bound the possible poses. The method, named MapAware Particle Filter, uses a nonlinear approach to mapmatching that can be integrated  into a particle filter framework for localization. Each particle is re-weighted based on the validity of its current position in the map. In addition, we buffer the trajectory followed by the vehicle and then append it to each particle’s pose. We  then quantify the overlap between the trajectory and the map’s free space. This serves as a measure of each particle’s validity given the trajectory and the shape of the map. We evaluated the method by performing experiments with different types of localization sensors: First, (i) we significantly reduced the drift inherent to dead reckoning. By only using wheel odometry and  map information we achieved loop closure over a distance of approximately 3 km. We also (ii) increased the accuracy of GPS localization. Finally, (iii) we fused a fragile 2D LiDAR localization with the map information. The resulting system had a higher robustness and managed to close the loop in an outdated map where it had failed before.

 

Submission to ICRA 2018 by KAdrian Rechy, Paulo V. K. Borges, Alberto Elfes, Andreas Pfrunder. Click here to download the paper.