Saugat Shahi

IMT Lucca

Tuesday, May 12, 2026, 11:00 - Friday, June 12, 2026, 12:00

SR 01-012

Abstract
Coordination of CAVs at an unsignalized intersection involves continuous trajectory optimization together with a combinatorial crossing order. Modeling the crossing order explicitly introduces discrete scheduling decisions and results in a mixed-integer nonlinear program (MINLP). As the number of vehicles grows, the resulting combinatorial complexity limits the feasibility of such formulations in real-time settings. We present a continuous optimization approach that avoids discrete variables by handling scheduling and control in a unified formulation through continuous entry and exit times and precedence constraints. A Lagrange-multiplier-based swapping heuristic is used to refine the crossing order: KKT multipliers associated with the precedence constraints identify critical adjacent pairs, and swaps are accepted only when they reduce the objective value. The method is implemented in two settings. The baseline uses CasADi with IPOPT in a receding-horizon loop. The real-time counterpart adopts acados with SQP-RTI and HPIPM, preserving the same problem structure while enabling substantially faster re-optimization. Current work focuses on eliminating swap re-solves entirely through parametric sensitivity analysis of the NLP solution, targeting a single RTI step per scheduling event.