Method and experiment for a hedgerow-type grape leaf-removing and harvesting robot

A hedgerow is one type of conservation buffer in grape cultivation. Mechanical harvesting has been widely used for the hedgerow-type grapes. However, it is often required to locate the peduncles of the grapes under the vines and leaves occlusion during harvesting. In this study, a leaf-removing mechanism was proposed to push away the obscuring vines, followed by the harvesting of the grape bunches. A leaf-removing harvesting robot was also developed using collaborative robotic arms. The efficient and low-damage harvesting of the grapes was achieved in complex occlusion scenarios. Firstly, a quantitative discrimination model was constructed for the peduncle visibility. The relative length, relative direction, and continuity were integrated to determine a visibility coefficient in the 0-1 interval. The occlusion degree of the peduncles was assessed in real time, including high, medium, and low visibility. In the grape peduncles with medium to low visibility, the optimal intervention point was identified to remove the occlusions by vines and leaves. Secondly, a spatial quadrilateral was constructed according to the endpoints of the grape peduncle and the occluding branch. The limited-memory broyden-fletcher-goldfarb-shanno was employed for the fermat-torricelli point as the optimal intervention point. Furthermore, a nonlinear mapping model was constructed from the end cartesian space to the joint space, in order to simplify the inverse solution of the robotic arm. The joint angle of the robotic arm was obtained corresponding to the end pose. The independent harvesting was achieved by the grape harvesting arm for the grapes with the highly visible peduncles, and the leaf-removing harvesting for the grapes with the medium to low visibility peduncles. Finally, the quantitative discrimination of the peduncle visibility was performed on 100 groups of samples. The results showed that the better performance was achieved in a visibility discrimination accuracy of 91.0% and a Kappa coefficient of 0.9. Among them, the discrimination accuracy for the high visibility was 94.1%. Field test results indicated that the harvesting damage rate was below 10.0% for the grapes with the highly visible peduncles, the success rate was 70.0%, and the average single-arm harvesting time was 3.2 s per cluster. While the occlusion handling mechanism was adopted in the grapes with the medium to low visibility peduncles. In the grapes with the medium visibility peduncles, the harvesting damage rate was below 23.3%, the success rate was 53.3%, and the average leaf-removing harvesting time was 8.7 s. In the grapes with the low visibility peduncles, the harvesting damage rate was below 36.6%, the success rate was 40.0%, and the average leaf-removing harvesting time was 14.8 s per cluster. The occlusion handling mechanism effectively distinguished the peduncles with the different occlusion degrees, and then switched the operating modes, indicating a relatively high harvesting success rate. The efficient, low-damage harvesting of the grapes was fully met under complex occlusion conditions in the hedgerow-type vineyards. The finding can also provide a strong reference for the mechanical harvesting of the hedgerow-type grapes.