Integrated Planning and Optimization Framework for Robotic Visual Inspection

Flexible, high-mix manufacturing demands inspection systems that plan not only for geometric coverage, but also for measurement quality and executable robot motion. This dissertation develops an integrated, model-based framework for discrete robotic visual inspection that couples viewpoint generation, quality-aware evaluation, kinematic feasibility, and trajectory optimization within a single planning pipeline. Starting from CAD-derived surface meshes and pinhole sensor models, viewpoint candidates are generated by direct surface sampling using Poisson-disc sampling, with hyperparameter tuning performed via Bayesian optimization. To move beyond the classical visibility matrix, the framework introduces a spatial-resolution-aware sampling density matrix that enforces target sampling density over non-uniform meshes. Moreover, a photometric visibility matrix is introduced that embeds material reflectance through Blinn-Phong and Cook-Torrance bidirectional reflectance distribution functions (BRDFs), excluding saturated or underexposed regions. Kinematic executability is addressed by a pose-refinement strategy that uses Bayesian optimization to transform infeasible inspection poses into feasible ones while preserving coverage properties. Planning is completed by formulating the set coverage generalized traveling salesman problem (SCGTSP) as an integer linear program that jointly selects one configuration per optimal inspection pose and orders the resulting tour under collision and workspace constraints. The approach is validated on synthetic and real experiments with 3-DOF, 6-DOF, and 8-DOF systems across diverse geometries and materials. Results show improved coverage with fewer optimal viewpoints, adherence to prescribed spatial resolution, and reduced inspection time compared to state-of-the-art methods that adopt sequential approaches. Our framework is sensor-agnostic and compatible with standard industrial toolchains, providing a rigorous and scalable foundation for automated quality inspection in modern production environments.