New traffic signal method reduces delays using regret-based control

New frameworks for traffic signal control reduce mean delay by up to 18.9% in real-world network tests.

New methods for traffic signal control can reduce mean delay in real-world networks. These methods use a Regret-Based Max-Pressure framework. This approach works even when traffic state information is imperfect.

Max-pressure (MP) traffic signal control selects phases using pressure scores. These scores come from estimated traffic states. Uncertainty in these states can change phase ranking or cause pressure loss.

Phase-selection regret is the pressure loss of a candidate phase compared to the pressure-maximizing phase. Stochastic Regret MP minimizes expected regret. Distributionally Robust (DR) Regret MP minimizes worst-case expected regret over a Wasserstein ambiguity set.

In tests on a 28-intersection real-world network, Stochastic Regret MP reduced mean delay by 18.9%. DR Regret MP reduced mean delay by 14.9%. These methods also work when detectors undercount or fail.

The methods reduce normalized true-state pressure regret by about 70% compared to Point-estimate MP. Robustifying regret makes the formulation invariant to state-dependent pressure components shared by all candidate phases.

Key facts

  • Stochastic Regret MP and DR Regret MP reduce mean delay by 18.9% and 14.9% respectively.
  • The methods reduce normalized true-state pressure regret by about 70% relative to Point-estimate MP.
  • The framework retains benefits during detector undercount and failure.

All sections

Edition