Abstract
Reliable robotic sorting and handling in food manufacturing remain challenging because foods vary in geometry, mechanical properties, surface condition, quality attributes, and spatial arrangement. These variations create coupled demands on perception, grasping, motion, control, and hygiene that component-level optimisation alone cannot resolve. This review examines grippers, perception and optical sensing, robotic platforms, and control strategies, together with their integration into food handling, sorting, packaging, and intralogistics. A recurring limitation is the gap between component performance and complete task success. Accurate recognition may not ensure a feasible grasp, while greater compliance can compromise force transmission or dynamic stability. Richer sensing can increase latency and integration burdens, and faster robot motion does not necessarily improve useful throughput. Based on the reviewed evidence, this review proposes an evidence-informed uncertainty-matched system-design perspective that links sensing, mechanics, control, and intelligence to product and process variability, dominant failure modes, and operating requirements. Industrial deployment requires validation across representative products, repeated operation, sanitation, failure recovery, and changeovers. Such task-level validation should assess complete systems for product integrity, hygienic compatibility, useful throughput, recoverability, and sustained reliability under realistic manufacturing conditions.