Development and field evaluation of a straddle-frame honeysuckle harvesting robot

Abstract

Honeysuckle (Lonicera japonica L.) holds significant economic value due to its traditional medicinal applications. However, the lack of modern mechanical harvesting methods renders the honeysuckle harvesting approach still manual. Increasing labour costs and a diminishing labour pool significantly increase the production cost of honeysuckles, potentially jeopardising sustainable production. Therefore, a straddle-frame honeysuckle-picking robot, which consists of a customised electric platform and gripper, was developed and field-tested. A rotatable two degrees-of-freedom gripping end-effector was designed to adapt to different honeysuckle orientations. The YOLOv8-pose algorithm was employed to detect picking points and ascertain flower orientation. This algorithm uses the keypoint located at the upper part of the honeysuckle as the picking point, thereby ensuring accurate positioning while simultaneously detecting the angle of the connecting line between the bud head and tail relative to the horizontal line and accordingly adjusting the rotation of the gripping finger. The model demonstrated robust performance in natural field environments, achieving 0.96 precision, 0.85 recall, 0.90 F₁-score, and 4.97° mean absolute error. Test results at various picking points indicated that the damage rate incurred when picking the honeysuckle in the bud's middle was about 20°%, while picking at the head resulted in minimal or no damage. For the picking performance evaluation experiment, a dataset of 2°06 honeysuckles was tested, resulting in a success rate of 66.8 %, a damage rate of 9.6°%, a positioning accuracy of 68.5°%, and average time to harvest one honeysuckle of 4.9 s. The developed honeysuckle harvest robot presents a promising solution for the automated harvesting of honeysuckle.