Robust vision–force fusion trajectory estimation for Ro-botic initial screw insertion under occlusion

Purpose Precision robotic initial screw insertion is frequently challenged by visual occlusions and dynamic force disturbances, which can degrade trajectory estimation accuracy and assembly reliability. Design/methodology/approach This paper proposes a dual-channel robust particle filter integrating visual and force feedback for re-al-time trajectory estimation. Stage-aware motion-continuity prediction, force-conditioned particle proposal, spatial attraction constraints, and Huber-based residual modelling are introduced to improve robustness under visual occlusion and uncertain contact. Findings Simulation and physical experiments show that the proposed method reduces 3D trajectory RMSE by 23.0% compared with single-channel particle filtering and by 27.6% compared with the extended Kalman filter, while improving insertion stability, contact compliance, and trajectory estimation accu-racy under visual occlusion and dynamic disturbances. Originality/value The proposed framework combines vision–force fusion, stage-aware particle propagation, and adap-tive robust estimation to provide an effective solution for high-precision robotic assembly under oc-clusion and uncertain contact conditions.