A small autonomous field robot for strawberry harvesting

Abstract

This paper presents the subsystem descriptions and testing of a small robotic platform intended to harvest strawberries that are grown on elevated beds in open field conditions. Agricultural robots can assist in the optimization of farm resources and help solve issues related to increasing farm costs and labor shortages. Monolithic, large size harvesters are already in development; however, they are susceptible to single-point-of-failure and lack the flexibility to adapt to varying field conditions and farm sizes. The proposed robotic harvester covers one row at a time and features a Delta arm configuration manipulator with a five-finger structure as the end-effector to individually pick strawberries. A deep neural network-based vision subsystem using a YOLOv4 model was adopted, which was configured to detect small objects and to locate and classify strawberries into five stages of maturity. During experiments on a commercial farm, the proposed platform, including vision, manipulation, and overbed navigation and control subsystems, achieved an overall success rate of 71.7% for five environment scenarios, with a minimum of 37.5% (the most complicated scenario) and a maximum of 94.0% (the easiest scenario). The average harvesting speed of the system was 7.5 s per strawberry.