Autonomous agricultural platforms require localization methods that remain reliable in orchard and agroforestry environments, where canopy, multipath, and intermittent satellite visibility can degrade Global Navigation Satellite Systems (GNSS) quality. This paper presents a low-infrastructure factor-graph fusion approach for robust agricultural localization, combining LiDAR-inertial odometry with raw GNSS observations and using Time-Differenced Carrier Phase (TDCP) constraints to improve drift resilience under changing field sensing conditions. The method operates at measurement level, preserves per-satellite signal-quality information, and is designed for practical field deployment without local reference infrastructure. Experiments on three agricultural traverses, spanning open-sky to dense-canopy conditions, show a 34% reduction in absolute positioning error and up to 39% reduction in relative positioning error in favorable conditions, while maintaining accuracy comparable to the LiDAR-inertial baseline and reducing relative drift by 6-14% under canopy. These results support consecutive-epoch TDCP as a practical localization aid for agricultural operations under non-stationary GNSS conditions in the field.
Towards Drift-Resilient Agricultural Localization via LiDAR-Inertial and Raw GNSS Factor-Graph Fusion
Osman, Anas
;Girlanda, Federico;Vecchio, Massimo;Antonelli, Fabio
2026-01-01
Abstract
Autonomous agricultural platforms require localization methods that remain reliable in orchard and agroforestry environments, where canopy, multipath, and intermittent satellite visibility can degrade Global Navigation Satellite Systems (GNSS) quality. This paper presents a low-infrastructure factor-graph fusion approach for robust agricultural localization, combining LiDAR-inertial odometry with raw GNSS observations and using Time-Differenced Carrier Phase (TDCP) constraints to improve drift resilience under changing field sensing conditions. The method operates at measurement level, preserves per-satellite signal-quality information, and is designed for practical field deployment without local reference infrastructure. Experiments on three agricultural traverses, spanning open-sky to dense-canopy conditions, show a 34% reduction in absolute positioning error and up to 39% reduction in relative positioning error in favorable conditions, while maintaining accuracy comparable to the LiDAR-inertial baseline and reducing relative drift by 6-14% under canopy. These results support consecutive-epoch TDCP as a practical localization aid for agricultural operations under non-stationary GNSS conditions in the field.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
