Picking path planning for one-eye-dual-arm green walnuts harvesting robot

To address the problems of long waiting time, limited coverage and suboptimal picking efficiency of single-arm picking robots when harvesting green walnuts in natural environments, this paper designs a picking scheme from "One-eye" to "dual-arm" for the green walnuts picking operations in unstructured agricultural environments, effectively expanding the picking range. In terms of visual recognition, we adopted the improved 3C-YOLOv8n network for training. We integrated the Residual Attention Block (RAB) into the C2f of the backbone network, introduced the Channel Attention Feature Map (CAFM) attention mechanism at the end, and incorporated the CA-HSPFN feature pyramid network structure into the neck network. After implementing the improvement strategy, the accuracy increased to 92.8%, the mean Average Precision (mAP) reached 94.5%, the model size decreased to 4.5MB, the precision improved by 1 percentage point, the mAP increased by 1.6 percentage points, and the model size reduced by 1.8MB. Asynchronous communication is adopted to mitigate the waiting time incurred during the idle phases of single-arm picking operations. An improved RRT algorithm with RDP-based path simplification was proposed for the dual-arm picking systems. Integrating the RDP algorithm cuts down path redundancy and trajectory points. The adaptive variable step-size strategy improves planning efficiency and shortens movement duration. Employing the preset initial configuration and master-slave arm coordination strategy achieves a improvement in success rate and reduction in average planning time. Verified by MATLAB® simulation, the average planning time of the improved RRT algorithm is reduced by 92.80% compared with the traditional method, the average path length is shortened by 10.07%, the average number of trajectory points is reduced by 81.39%, and the average number of iterations is reduced by 78.21%. Under laboratory conditions, physical experiments were systematically carried out utilizing the JAKA dual - arm system. The picking efficiency of the proposed scheme was measured at 16.11 seconds per item. When compared to the traditional single - arm scheme, this represented a 27.69% increase in efficiency.