Development of a stereo vision-based UGV guidance system for bareroot forest nurseries

The US forest nursery industry still relies heavily on manual labor for inventories during the growing season. Automating these processes could significantly reduce labor requirements while enhancing efficiency in counting and quality assessment of bareroot forest tree seedlings. In this study, we developed a stereo vision-based automatic guidance system for an unmanned ground vehicle (UGV) designed to follow forest nursery beds. The system was first developed and validated in simulation using the Webots simulator integrated with the Robot Operating System (ROS). A depth image processing algorithm based on the Hough transform was employed to detect bed edge-lines, with the image-derived relative bed angle serving as the error signal for a proportional controller that corrected the vehicle's heading and lateral deviation relative to the bed. The physical system was subsequently tested on a skid-steering UGV at a commercial forest nursery across four curved pine seedling beds with varying radii and lengths under an overcast condition, at ground speeds of 0.5 m/s and 1.0 m/s. The proposed vision guidance system successfully completed all eight trials without collisions with seedlings, traversing a total distance of 1,422 m while achieving an average root mean square deviation of 0.05 m. These results underscore the potential of stereo vision-based guidance for automating nursery operations, paving the way for more efficient and reliable inventory and quality assessment in the forest nursery industry. Future research should validate and improve the system performance across a variety of bed geometries, soil characteristics, pine seedling heights, lighting conditions, and speeds.