Saugat Shahi

IMT Lucca

Tuesday, October 27, 2026, 11:00 - 11:59

Room 02-012, Georges-Köhler Allee 102, Freiburg 79110, Germany

Optimizing the crossing order of connected and automated vehicles (CAVs) at an intersection requires evaluating the trajectory optimization problem for many candidate orders, which is too expensive for real-time operation when it is solved as a single centralized NLP. We exploit the structure of the problem in bi-level form: each vehicle solves its own optimal control problem, parametrized by its entry and exit times, and an upper-level problem optimizes these times. First- and second-order derivatives of the resulting value functions follow from parametric sensitivity analysis of the lower-level KKT systems. They provide the curvature for an SQP method and a local quadratic model that predicts the effect of reordering vehicles, enabling decentralized coordination in real time.