Fusion‐Odometry‐Based Global Navigation in Closed Orchards

ABSTRACT Autonomous navigation in closed orchards is essential for agricultural production, yet robust global localization remains challenging due to Global Navigation Satellite System (GNSS) obstruction. This study proposes a fusion‐odometry (FUO)‐based global navigation framework that integrates localization, path planning, and path following. FUO employs real‐time kinematic odometry (RTO)‐based initialization to ensure global consistency. It refines 2D LiDAR–Inertial Odometry via Smoothing and Mapping with drift correction. Transformed Wheel Odometry in the global frame (WHOT) and Inertial Measurement Unit data are then fused through an Extended Kalman Filter to provide lightweight and globally consistent odometry. Finally, path tracking is performed with an improved pure‐pursuit controller. Field experiments demonstrated the effectiveness of the framework across diverse conditions: in open‐field benchmarks, FUO achieved centimeter‐level accuracy and maintained loop closure at ground‐truth quality, whereas WHOT appeared cumulative drift; in canopy‐covered crabapple orchards, FUO mitigated the impact of RTO failures under GNSS occlusion and achieved substantially smaller loop‐closure errors than RTO; and in a dormant tall‐spindle apple orchard, FUO further confirmed cross‐scenario and seasonal generalization, with absolute path tracking errors consistently below 0.06 m and 4.23°. In summary, the proposed framework contributes both a conceptual advance in multisource odometry fusion and a practical navigation solution tailored to orchard environments, substantially outperforming individual odometry modules and enabling deployment‐friendly autonomous operations under GNSS‐challenged conditions.