Autonomous Weed Mowing Based on LiDAR-SLAM

Abstract

　This study aimed to achieve autonomous operation of an electric mower in global navigation satellite system–denied environments. We demonstrated autonomous weed mowing using LiDAR-SLAM with 3D-LiDAR in a mountainous farmland. Based on self-localization through NDT matching between a prebuilt 3D point-cloud map and real-time scanned point cloud, the mower accurately followed the entire planned route in the farmland. Within the target area, the root mean square cross-track error was 7 cm, and the uncut weed area was less than 5 %. Additionally, using a point-cloud map generated after mowing, autonomous mowing was successfully achieved via self-localization with NDT matching, even when the grass had regrown.