Design and evaluation of a multi-sensor fusion navigation system for autonomous orchard weeding robot

Autonomous navigation is a fundamental requirement for intelligent orchard management and the alleviation of agricultural labor shortages. However, conventional GNSS-based navigation systems often suffer from severe performance degradation under canopy occlusion, while single-sensor approaches exhibit limited robustness in complex and hilly orchard environments. To address these challenges, this study proposes a low-cost and robust multi-sensor fusion SLAM-based navigation system for orchard weeding robots. The proposed system integrates RGB-D camera–based point cloud information with 2D LiDAR data to construct high-precision two-dimensional grid maps, enabling reliable representation of complex orchard environments. To mitigate feature degradation in unstructured scenes, a multi-sensor fusion localization framework is developed by tightly integrating visual–inertial odometry (VIO) with wheel odometry. In addition, Adaptive Monte Carlo Localization (AMCL) is introduced to enhance global positioning stability, while Cartographer enhanced by multi-sensor fusion localization is employed to improve mapping accuracy and long-term consistency. Field experiments conducted in a real orchard environment demonstrate that the proposed system achieves an average absolute pose error of 0.486 m with a standard deviation of 0.328 m. Furthermore, the navigation system satisfies the accuracy requirements for autonomous weeding operations, maintaining an average heading deviation below 7°, and longitudinal and lateral positioning errors within 8 cm and 7 cm, respectively, at operating speeds ranging from 0.5 to 1.5 m/s. The experimental results verify that the proposed approach provides a practical and reliable solution for GNSS-denied orchard navigation, showing strong potential for improving the efficiency and sustainability of intelligent orchard management.