Abstract

For an intelligent agricultural robot to reliably operate on a large-scale farm, it is crucial to accurately estimate its pose. In large outdoor environments, 3D LiDAR is a preferred sensor. Urban and agricultural scenarios are characteristically different, where the latter contains many poorly defined objects such as grass and trees with leaves that will generate noisy sensor signals. While state-of-the-art methods of state estimation using LiDAR, such as LiDAR odometry and mapping (LOAM), work well in urban scenarios, they will fail in the agricultural domain. Hence, we propose a mapping and localization system to cope with challenging agricultural scenarios. Our system maintains a high quality global map for subsequent reuses of relocalization or motion planning. This is beneficial as we avoid the unnecessary repetitively mapping process. Our experimental results show that we achieve comparable or better performance in state estimation, localization, and map quality when compared to LOAM.