A staged vision-force collaborative framework for precision robotic insertion of metallic valve components

Precision robotic insertion of metallic valve components remains challenging because specular reflection, depth loss, and partial occlusion degrade visual localization near the hole entrance, while contact further reduces the reliability of vision. The resulting residual pose errors often cause impact, side loading, jamming, and insertion failure. To address this issue, we propose a staged vision–force collaborative framework (SVFC) for precision insertion of metallic valve components. The key idea of SVFC is to treat metallic insertion as a sensing-reliability handover process: vision is used to compress the global pose error into a force-recoverable local neighborhood, while force feedback subsequently takes over for residual centering and contact regulation. The framework divides the process into three sequential phases: vision-guided coarse alignment, force-guided residual pose correction, and compliant insertion under variable-parameter admittance control. During residual correction, the joint held by the gripper is driven along a local Archimedean spiral trajectory within the vision-defined neighborhood, and centering is determined using a convergence criterion based on the root-mean-square values of lateral forces and torques. During insertion, virtual stiffness and damping are adjusted online according to contact state to suppress transient impact and mitigate jamming under different fit conditions. Experiments on representative clearance-fit and interference-fit tasks show that the proposed method improves insertion success rate, reduces insertion time, and lowers peak contact forces compared with baseline methods, indicating that explicit coordination between vision-dominant alignment and force-dominant contact regulation is beneficial for metallic insertion.