ALOT: Augmented Localization with Obstacle Tracking

Zhikai Li, Krittin Kawkeeree, Yue Linn Chong, Christina Dao Wen Lee, Marcelo H. Ang

Abstract: Augmented Localization with Obstacle Tracking (ALOT) is a pipeline for localization, involving closed-loop feedback between an obstacle tracker and a particle filter localization. The tracker tracks and labels dynamic obstacles it sees and uses historic information to predict positions of dynamic obstacles at the current time-step. Following up on this, the tracker uses the current observation along with predicted obstacle positions to proposes ego poses for localization. The localization method in ALOT employs a particle filter. During scan matching, it removes dynamic obstacles from the scan using information obtained from the tracker. Particles are weighted once during scan matching, and a second time with ego-pose proposals provided by the tracker. Upon reconstructing the ego-pose belief, the particle filter localization provides a feedback to the tracker with the most likely ego-pose to allow the tracker to update its tracking and further propose ego-poses at the next time-step. ALOT is tested on real-world data collected in a laboratory. In low to moderately dynamic environments, it achieves an average positional and heading errors of 0.171 m and 1.63 $$^\circ $$ respectively. When run in larger crowds, ALOT has positional and heading errors of 0.467 m and 4.784 $$^\circ $$ .