How Self-Driving Cars Could Quietly Create Massive Congestion Through Ghost Trips

The study warns that autonomous vehicles could worsen congestion through empty “ghost trips” unless cities redesign parking supply and pricing. Using real-network simulations, researchers show that a strategic mix of parking construction and pricing can significantly reduce these impacts as AV adoption grows.

Researchers from the University of Illinois Chicago, Sharif University of Technology, and the University of Minnesota have taken an early look at a future problem that cities have barely begun to prepare for: how self-driving cars will reshape congestion once they start dropping off passengers and heading off alone to park. Their study shows that autonomous vehicles (AVs), freed from the need to keep drivers close to their destinations, will generate large numbers of empty “ghost trips” as they cruise to cheaper or more available parking. These extra miles could turn into one of the most significant new sources of urban congestion, unless cities rethink how parking supply and pricing are designed.

A Real Network, Not a Thought Experiment

Many earlier studies relied on simplified bottleneck models or hypothetical monocentric cities. This research instead embeds AV behavior into the real Sioux Falls traffic network, capturing its CBD, midtown and suburban dynamics. Using the SUMO mesoscopic simulator paired with a genetic optimization algorithm, the team tested thousands of parking-supply and parking-price combinations. Each AV in the model decides where to park by weighing round-trip travel distance and parking cost, then the simulator routes all trips dynamically through morning and evening peaks. This approach captures the complex timing dependencies between passenger trips and AV repositioning, offering a much more realistic view of how congestion unfolds.

Why Pricing Alone Fails

One takeaway is immediate: pricing by itself cannot fix the congestion caused by AV repositioning. Even well-calibrated prices cannot overcome a simple constraint, if a zone lacks enough parking supply, AVs will spill over into other areas, often creating additional congestion. The genetic algorithm repeatedly found that the best outcomes came from jointly optimizing where parking is built and how it is priced. Compared with random allocations or incremental expansions of existing parking, the optimized solutions sharply reduced network travel times, in some cases performing nearly as well as the theoretical best case where every AV parks at its destination with no repositioning.

The Geography of Tomorrow’s Parking

A striking pattern emerges when looking at different levels of AV adoption. At low penetration rates such as 20 percent, midtown zones offer the best balance of land cost and accessibility. Concentrating parking there minimizes ghost miles without the expense of building large new structures in the CBD. But as AV adoption rises to 40 or 50 percent, the calculus flips. The extra repositioning traffic makes it increasingly valuable to build more parking directly in the CBD despite higher construction costs. Reducing travel distance for empty AVs becomes more important than saving money on land. The study also shows that at higher AV penetration, the network becomes more sensitive to investment: each added dollar of parking budget produces greater travel-time improvements compared to lower penetration levels.

Planning Now for the AV Era

The researchers also highlight the importance of allowing AVs to “park at home.” When this option is included, the model can allocate budgets more strategically instead of being forced to build enough spaces for all AVs. Without home parking, every scenario becomes unrealistically expensive and less efficient. Perhaps the clearest warning for city planners comes from the comparison between optimized and random parking strategies. Randomly scattering parking or setting prices without coordination produces results barely better than providing no parking at all. Built spaces remain unused, attractive zones remain undersupplied, and ghost trips multiply. By contrast, a targeted, data-driven combination of supply and pricing substantially reduces congestion even with limited budgets.

Taken together, the findings send a strong message: AVs will not automatically ease urban traffic. If left unmanaged, their freedom to cruise empty could worsen peak-hour congestion dramatically. But with strategic planning, rooted in simulation, sensitivity analysis, and carefully calibrated pricing, cities can harness AV mobility while protecting network performance. The window to prepare is open now, long before fleets of empty AVs begin shaping the daily pulse of rush hour.