A recently proposed coordination method quickly improves the order in which vehicles cross an unsignalized intersection. The scheme only runs a nonlinear program (NLP) when a new vehicle arrives, adding the newcomer to the end of the current crossing sequence and then testing a few nearby swaps. The cost sensitivities from the NLP guide which swaps are likely to reduce travel‑time and control effort. A memory of previously evaluated swaps keeps the computation small across events.

Simulations with hundreds of heterogeneous cars show that this approach recovers more than 85 % of the gap left by a simple first‑in‑first‑out policy. For streams of 301 vehicles the method cuts average cost by 51 % in sparse traffic and 67 % in dense traffic, while keeping the number of NLP solves per arrival modest.

The results confirm that an arrival‑triggered receding horizon can bring real‑time unsignalized intersection control close to optimal, but the approach still requires a full NLP for each candidate swap, which limits scalability for very high vehicle counts.