自動運転車は渋滞を減らすはずだった——もし減らなかったら？ Autonomous vehicles were supposed to cut traffic—what if they don't?

For years, autonomous vehicles have been pitched as a cure for urban gridlock. The logic seemed sound: remove human error, optimize routing with AI, and traffic should flow more smoothly. A new study covered by Ars Technica complicates that narrative significantly.

Data on Waymo's robotaxi operations reveals that the vehicles travel empty for nearly half of all miles driven. In industry parlance, these are "deadhead miles" — the stretches between dropping off one passenger and picking up the next, or repositioning toward high-demand zones. The study concludes that robotaxis, at least in their current form, don't meaningfully reduce traffic compared to conventional ride-hailing services like Uber or Lyft.

Deadhead miles have been a known problem since ride-hailing went mainstream. Research published around 2019 estimated that ride-hailing vehicles were responsible for as much as 60 percent of increased congestion in cities like San Francisco — not because more people were traveling, but because the cars themselves were circulating constantly in search of fares. The hope was that autonomous vehicles, freed from human scheduling constraints and guided by smarter dispatch algorithms, would crack this problem. The new data suggests that hasn't happened yet.

There's a structural reason why. Robotaxis still need to be where riders are, and riders are not always conveniently located near where the last trip ended. Without a passenger on board, the vehicle has to move anyway. In some respects, the economics of autonomous operation could make things worse: without driver wages to worry about, fleet operators face fewer cost penalties for running vehicles empty, potentially incentivizing larger fleets and more deadhead movement.

Waymo and other players in the space are likely to contest some of these framing points. The company has emphasized safety records and per-passenger emissions reductions as primary metrics of success. There's also a reasonable argument that the technology is still early-stage — if pooled or shared rides become the norm rather than the exception, utilization rates could improve substantially. Currently, solo rides dominate robotaxi usage, and that's where much of the inefficiency lives.

From a urban planning perspective, the findings reinforce a growing consensus that autonomous vehicles are not a standalone solution to traffic. Policymakers in cities where robotaxis operate — San Francisco, Phoenix, Austin — are increasingly asking how these services integrate with buses, rail, and bike infrastructure rather than replace them. A robotaxi that gets someone from a transit hub to their front door is a very different proposition than one competing directly with a bus route.

The broader takeaway may be that technological optimism around AVs needs to be paired with harder questions about deployment models and city design. The vehicles may be autonomous, but the traffic problems they were supposed to solve are deeply human in origin — shaped by sprawl, zoning, and commuting habits that no routing algorithm changes on its own.