A GNSS-free LiDAR-based navigation architecture for autonomous inter-row operation under sparse or absent vegetation conditions

Autonomous navigation in agricultural inter-rows is challenging under sparse or absent vegetation conditions, when the only reliable landmarks are vineyard posts, and systems based on the Global Navigation Satellite System (GNSS) are unstable or unavailable. This study presents and validates in the field a mapless approach based on Light Detection and Ranging (LiDAR) sensor and robust RANdom SAmple Consensus (RANSAC) regression. This approach exploits LiDAR returns from vineyard posts and estimates the central inter-row line, from which lateral offsets and rover orientation (yaw) are derived, without satellite dependencies. The performance of the proposed approach was evaluated in the field in a vertically trained vineyard and in a vineyard simulation scenario, considering the measured lateral error, robustness to outliers and partial occlusions caused by posts reflections and missing LiDAR returns, and mean processing time. In the real vineyard scenario, a mean lateral error of 0.094 m with a standard deviation of 0.135 m was observed, with mean processing time of 119 ms. In the vineyard simulation scenario, the obtained mean lateral error was 0.071 m, with a standard deviation of 0.124 m and mean processing time of 444 ms per frame. The results demonstrate the feasibility of inter-row autonomous guidance based exclusively on the arrangement of vineyard posts, without reliance on satellite positioning even in absence of vegetation.