GEERS: Georeferenced Enhanced EKF Using Point Cloud Registration and Segmentation
Rui Bettencourt, John Lewis, Rodrigo Serra, Meysam Basiri, Alberto Vale, Pedro U. Lima
Abstract
Georeferenced Enhanced EKF using point cloud Registration and Segmentation (GEERS) is a high-accuracy and consistent-rate localization method for outdoor robots. The localization is estimated by an EKF that fuses wheel odometry, IMU and GNSS measurements, in addition to feedback corrections from a registration step. The method improves localization accuracy by registering range sensors with pre-obtained georeferenced 3D maps and providing feedback corrections to the EKF. The continuous fusion of GNSS measurements naturally provides an initial estimate and reduces kidnapped robot situations in symmetric environments. The proposed method can integrate any range sensor (such as RBG-D cameras or 2D and 3D LiDAR). Experimental results in a real-world solar farm, its simulated digital twin, and an open dataset demonstrate localization accuracy improvements. Real-world experiments on a solar farm demonstrated the flexibility and reliability of the proposed method, exposing its advantages towards GNSS-only-based approaches.
BibTeX
@inproceedings{ral2024_geersgeoreferenc,
title = {GEERS: Georeferenced Enhanced EKF Using Point Cloud Registration and Segmentation},
author = {Rui Bettencourt and John Lewis and Rodrigo Serra and Meysam Basiri and Alberto Vale and Pedro U. Lima},
booktitle = {RA-L 2024},
year = {2024}
}