Mapping and localization methods for water and fertilizer inspection robots in facility blueberry cultivation on unstructured terrain

Abstract To address the challenges of high system costs associated with the reliance on multi-node fixed sensors in traditional water and fertilizer monitoring, as well as mapping distortion and limited localization accuracy of the robot on unstructured terrain, this paper designs a track-type robot for facility blueberry cultivation and proposes corresponding mapping and localization methods. The robot integrates key modules such as two-dimensional LiDAR, soil sensors, and a robotic arm, enabling autonomous movement to replace the traditional multi-node fixed monitoring approach for the collection and inspection of water and fertilizer data. To address the point cloud distortion caused by unstructured terrain, a pose-adaptive cascaded point cloud correction method is introduced. This method improves point cloud quality and mapping accuracy through sensor data preprocessing, motion distortion compensation, and pose distortion compensation. For localization, a combination of EKF state fusion and AMCL localization strategies is employed to mitigate the impact of track slip and ground undulation on positioning accuracy. Experimental results show that the maximum mapping error of the robot is 0.021 m, with localization errors of 0.025 m, 0.034 m, and 0.048 m at different distances. The overall localization error is controlled within ±8 cm. This robot provides a cost-effective solution for water and fertilizer detection in facility blueberry cultivation and offers valuable technical support for precise water and fertilizer management.