Advancement and field evaluation of a honeysuckle harvesting robot with integrated clamping and air-suction mechanisms

Abstract

Honeysuckle is a high-value economic crop, currently relies entirely on manual labor for harvesting, making the process labor-intensive and costly. Existing research studies, however, face challenges such as low throughput and poor harvest performance, as well as a lack of an efficient collection system for harvested honeysuckles. Therefore, this study presents the advancement and field evaluation of a honeysuckle harvesting robot integrated with clamping and air-suction collection mechanisms, aimed at improving harvesting performance and operational stability in real-world environments. The system integrates an RGB-D camera, a 3-DoF linear manipulator, a 2-DoF (rotating and clamping) end-effector, an air suction collection mechanism, and a crawler chassis to enable continuous clamping, separation, and collection of honeysuckle. Compared to previous designs, the robot system has been improved in both software and hardware aspects. On the software side, an advanced system was developed by integrating an enhanced YOLOv8-based detector with an improved ant colony optimization algorithm for honeysuckle picking point detection, orientation estimation, and continuous path planning. Regarding the hardware, an air suction collection system was developed, which directly combined the air suction with the gripper to achieve direct non-destructive collection after separation for continuous harvesting. Field tests were conducted in a commercial honeysuckle field under standardized planting in Julu County, Hebei Province, China. Its performance on 352 mature buds at various densities was evaluated, and the results showed a detection accuracy of 96.3 %. The separation success rate was 76.1 %, which was the major reason for low overall harvest success rate to 68.2 %. The field test also demonstrated that the air suction success rate was 93.1 %, and the damage rate was 6.5 % (all rates are macro-averaged). The average harvesting cycle time was 3.4 s/mature bud, which included only the picking and collection operations within a single plant and did not include traveling time between plants. Compared to experimental results already reported, the improved harvesting robot reduced the harvesting path length by 34 % and harvesting cycle time by 28 %. The improved system shows good potential for fully autonomous honeysuckle harvesting.