Robust Autonomous Navigation in Vineyard for complete path coverage

Autonomous navigation in agricultural environments has emerged as a key innovation in modern agritech, providing significant benefits in terms of efficiency, cost reduction, and sustainability. In particular, autonomous systems in vineyards can alleviate labor shortages, optimize resource usage, and enable precision agriculture practices. However, the structured yet highly variable nature of vineyards presents unique challenges for autonomous navigation. Vineyards are characterized by narrow and uneven pathways, curved and sloped terrains, and dense vegetation that can obstruct visibility and interfere with traditional localization methods. Furthermore, environmental factors such as changing lighting conditions, seasonal variations, and occlusions from foliage add additional layers of complexity to perception and decision-making systems. To achieve reliable autonomy, a combination of advanced perception, localization, and control strategies is necessary. Traditional approaches primarily rely on GNSS-based localization, often complemented by LiDAR, IMU, and wheel odometry to enhance accuracy and robustness. While these sensor fusion techniques have proven effective in open-field agricultural applications, they encounter significant limitations in vineyards and orchards. Tall and dense vegetation can obstruct GNSS signals, leading to localization errors and reduced reliability in environments with limited satellite visibility. Additionally, wheel odometry can suffer from drift and inaccuracies on uneven or slippery terrain, further complicating long-term navigation. This thesis explores both localization-based and position-agnostic solutions to address these challenges. By leveraging behavior trees, an approach widely used in robotic control and decision-making, the proposed navigation system benefits from modularity, hierarchical structure, and real-time feedback mechanisms. These characteristics enable flexible and adaptable control pipelines capable of handling complex vineyard environments with varying terrain and occlusion conditions. The developed solutions are extensively evaluated through a combination of simulated environments and real-world experiments, providing a comprehensive assessment of their performance, robustness, and applicability in practical agricultural settings.