Abstract: To address the issue of occlusion-induced model uncertainty during the assembly of industrial robotic arms,which complicates the accurate estimation of hole part poses,an improved method combining the YOLOv8 model with adaptive unscented kalman filtering,named SHUK-SVSF algorithm,was proposed. Firstly,to reduce response loss caused by occlusion and improve hole part pose detection accuracy,the SEAM module was introduced into the output layer of YOLOv8 model′s neck. Secondly,a design based on the Sage-Husa algorithm adaptive unscented kalman filter was proposed,which adjusts the measurement noise covariance matrix in real-time,thereby enhancing the stability and accuracy of pose estimation in dynamic environments. Finally,the introduction of variable structure gain within the unscented kalman filtering leverages the precision of unscented kalman filtering and the stability of smooth variable structure,further improving pose estimation performance. Experimental results demonstrate that the modified YOLOv8 method exhibits high detection accuracy in occluded scenarios,while the proposed SHUK-SVSF algorithm meets the real-time requirements for visual guidance in robotic arm assembly processes and achieves high accuracy and robustness in pose estimation under model uncertainty conditions.

Key words: industrial robotic arm, adaptive unscented kalman filtering, occlusion response loss, hole parts, pose estimation