Vision-Guided Dual-Arm Robotic Apple Harvesting System

Abstract

To address the issues of high labor intensity and low efficiency in traditional manual picking methods, this study designs a vision-guided dual-arm apple picking robot tailored for dwarf-type and dense-planting orchards. The robot features a symmetrical configuration with dual six-degree-of-freedom collaborative robotic arms and a hierarchical dual-camera perception system. For visual target detection, an off-the-shelf YOLOv8 model is utilized to rapidly provide target regions under severe canopy occlusions. Furthermore, precise 3D spatial localization is realized via HSV-Otsu segmentation and point cloud radius outlier removal. To ensure efficient and collision-free dualarm coordination, a task allocation strategy based on spatial partitioning is implemented, where each arm’s target queue is prioritized by depth (Z-axis, closest first). Combined with Rapidly-exploring Random Trees (RRT) for real-time motion planning, experimental results demonstrate that the system operates without physical interference, achieving an 87.87% picking success rate and reducing the average cycle time per fruit by over 35%. This research provides a highly efficient and feasible solution for automated orchard picking.