LiDAR-based Structure Tracking for Agricultural Robots: Application to Autonomous Navigation in Vineyards

Autonomous navigation is a key defining feature that allows agricultural robots to perform automated farming tasks. Global navigation satellite system (GNSS) technology is providing autonomous navigation solutions for current commercial robotic platforms that can achieve centimeter-level accuracy when real-time kinematic (RTK) corrections are available. However, GNSS-based solutions are expensive and require a long preparation phase where the field has to be surveyed with a GNSS rover to collect waypoints for the navigation path. An alternative navigation approach can be provided by Local perception sensors, such as LiDAR scanners, by tracking geometric features in the perceived scene. This paper presents a robust LiDAR-based solution for structure tracking along vine rows. The proposed method does not require prior field surveying, and it is insensitive to crop characteristics such as row width and spacing. Moreover, the proposed algorithm identifies and builds an online regression model of the structure. This is done by applying the Hough transform with a parameterization and search method motivated by a practical interpretation of point cloud statistics. The proposed method was tested on a commercial robotic platform in two configurations of vineyards. The experiments show that the proposed algorithm achieves consistent and accurate row tracking, which was validated against a reliable RTK-GNSS ground truth.