The success of inertial confinement fusion experiments hinges on the production of perfectly round spherical capsules placed at the center of an implosion. Some of the most common ablator materials are grown on poly(alpha-methylstyrene) (PAMS) mandrels. Human operator–based optical inspection of individual PAMS mandrels followed by a selection decision, is a labor-intensive process that suffers from operator dependence. General Atomics has developed a robotic system to handle and image these delicate PAMS mandrels and has implemented an autonomous method for evaluating shell quality. The selection criteria of acceptable mandrels has been standardized by employing visual defect characterization tools and associated machine learning algorithms. This work discusses the mechanical upgrades made to the robot cell for handling shells, the suite of software tools developed for a more complete evaluation of individual shells, and correlating defect statistics from entire batches to production data from the PAMS fabrication process parameters.