Pick-to-Learn Calibrates an MPC Policy for Origin-to-Destination Flight Planning

arXiv·low signal

Marco C. Campi and Simone Garatti (arXiv 2607.16084, eess.SY/cs.LG) apply the Pick-to-Learn methodology — originally developed for machine learning generalization bounds — to calibrating a Model Predictive Control policy on a flight routing problem. The transfer is the point: a technique built to certify learned models is repurposed to give a classical controller data-driven tuning with retained guarantees. Narrow application, but the pattern of importing ML certification machinery into control is worth tracking.