Navigation System for Autonomous Agricultural Vehicles for Indoor Farms

Simultaneous Localization and Mapping (SLAM) is a fundamental technology in autonomous driving and refers to realizing autonomation and intelligent navigation functions. The SLAM is used to integrate localization function with the construction of the surrounding environment. When the robot vehicle moves, the system builds a consistent map using sensing of the environment according to its pose estimation. This research aims to design an autonomous agriculture vehicle for scouting based on stereo visual SLAM in the indoor environment. This research utilized a stereo camera as the primary sensor to perceive surrounding environment information. The system extracted image features between each continuous frame and then regards these features as natural landmarks. As a result, the system could estimate the pose and movement between these frames and updated the global map, which was also a whole process of SLAM. This research outcomes reported image local invariant features extracting algorithm, global localization, environment mapping based on the image frames, path planning, and path tracking. The ORB_SLAM was used, which is one of the SLAM repositories that oriented FAST and Rotated BRIEF to describe features. Our research used the ORB method to realize the SLAM function and compared the SIFT and ORB methods in different environments. The result shows that when these two methods analyze the same images, the ORB method exhibited excellent performance and satisfied the need for real-time operation with higher accuracy. The fast library for approximate nearest neighbors (FLANN) matching algorithm had excellent matching results and ensured that the estimated pose was right for generating accurate global maps. Robot Operation System (ROS) is an open-source meta-operating system for robots that was used to construct the whole navigation system part. The objective of the navigation section was to navigate the vehicle to the designated destination. This research’s path planning and tracking algorithm used Dynamic-Window Approach (DWA) and pure-pursuit, which updated the map in real time. According to the dynamic map, the system demonstrated a no-collision path to the destination.