A Cost-Effective Global Localization System For Precision Agriculture Tasks In Polytunnels

Precision agriculture tasks demand highly efficient and accurate actions performed by autonomous robots. In order to carry out such actions, a prerequisite for a robot is to accurately localize itself in its working environments, such as crop fields, greenhouses, polytunnels, etc. Agricultural environments usually present unique challenges to localization tasks such as the highly repetitive structure of a polytunnel or a crop field leading to visual aliasing and changing appearance over time. This makes it challenging for the robot to localize. In this paper, we develop a 6DoF localization system for precision agriculture tasks in polytunnels. The system only requires a cost-effective stereo RGB-D camera and a prebuilt 3D map. The system allows an agricultural robot to robustly localize over multiple stages of a strawberry season despite the strong dynamic and changes in the appearance of the environment caused by growing plants. Experiments are carried out on a real strawberry polytunnel over a period of several weeks to evaluate the system. The results show that our system provides adequate localization accuracy for agricultural robots to perform tasks that require a high level of accuracy.