Robot-assisted multiview collaborative measurement and planning for complex-surface reconstruction

Abstract

Complex free-form components contain strong curvature variations, deep occlusions, and narrow passages that cannot be reconstructed reliably from a single structured-light view. This paper presents a robot-assisted multi-view measurement and planning workflow using an eye-in-hand Mech-Eye Nano 3D camera and a UR manipulator. The camera is modeled as a factory-calibrated measurement unit, while external view generation, visibility testing, hand-eye transformation, collision-aware execution, and point-cloud fusion are formulated in a common part coordinate system. Candidate views are generated from the CAD surface and ranked by a discrete patch-wise utility that combines newly observable area, incidence quality, redundant coverage, and robot motion cost. A nine-blade impeller was measured with seven automatically selected primary views. Five independent hand-eye validation groups produced a mean translation error of 0.537 mm and a maximum error of 0.930 mm. The seven-view fusion contained 2,401,835 valid points and covered 84.155% of 20,000 CAD samples at a 5 mm completeness threshold; the curvature-weighted coverage was 25.932%. The final view added 1.905 percentage points of ordinary coverage, confirming diminishing returns and motivating closed-loop local rescanning. The reported coverage is a completeness measure rather than dimensional accuracy.