Researchers Use AI to Cut Highway Congestion via Connected Vehicles

Congestion on highways costs time and fuel, and researchers have long sought automated solutions. A team has now developed a system that uses reinforcement learning to manage traffic flow by controlling the spacing between connected vehicles—a strategy that outperforms both traditional speed-limit signs and existing vehicle-control methods.

The Problem with Current Approaches#

Today's highways typically rely on variable speed limit (VSL) signs to manage congestion, but these suffer from a critical flaw: they update infrequently and drivers often ignore them. More recent research has explored controlling individual vehicles directly, but these "Lagrangian" strategies require knowing when drivers plan to change lanes—information that is often hidden or uncertain, making reliable control difficult in real multi-lane traffic.

A New Strategy: Headway Control#

The researchers propose an alternative Eulerian control system that sidesteps the lane-change problem by focusing on aggregate traffic density near bottlenecks. Rather than telling vehicles how fast to go, the system issues headway commands—instructions about the distance or time gap that should be maintained between consecutive vehicles. This approach leverages connected automated vehicles (CAVs) equipped with adaptive cruise control (ACC), which can follow such instructions reliably.

The team tested two variants: one using time-headway (a set duration between vehicles) and another using distance-headway (a fixed gap in meters). Both were optimized through reinforcement learning to find the best headway values under different traffic conditions.

Results and Validation#

In large-scale simulations, the system achieved significant improvements. It boosted traffic flow by up to 10.6% over uncontrolled human traffic and 6.7% better than traditional VSL approaches. The researchers also developed a new evaluation metric to address a common flaw in traffic simulations: accounting for the boundary effects that occur when vehicles continuously enter and exit the system.

Why It Matters#

The emphasis on headway rather than speed control makes the approach more practical for real-world deployment. It avoids the complexity of tracking hidden driver intentions and works naturally with existing ACC technology already present in many vehicles. The work suggests a deployable path toward safer, more scalable congestion mitigation on highways.