Autonomous vehicles were supposed to cut traffic—what if they don’t?

自动驾驶汽车本应减少交通拥堵——如果它们做不到呢？

The age of robotaxis, long the preserve of science fiction, is now a reality, at least in a handful of American cities. It took just over a decade to get from the DARPA Grand Challenges to the start of Waymo’s commercial service in California, albeit initially with a safety driver on board. Proponents of the technology, which has attracted at least $100 billion in investment, say robotaxis will be safer than human-driven vehicles. And last year, Waymo’s data showed its cars were involved in many fewer crashes than human drivers, with much lower insurance claims, although recent issues with school buses and flooded roads show the technology isn’t perfect.

自动驾驶出租车（Robotaxi）的时代曾长期存在于科幻小说中，如今已成为现实，至少在少数几个美国城市是这样。从 DARPA 挑战赛到 Waymo 在加州开启商业化服务，仅用了十年多一点的时间，尽管最初车内还配备了安全员。这项技术吸引了至少 1000 亿美元的投资，其支持者认为自动驾驶出租车将比人类驾驶的车辆更安全。去年，Waymo 的数据显示，其车辆涉及的碰撞事故远少于人类驾驶员，保险索赔也低得多，尽管最近在校车和洪水道路上出现的问题表明该技术尚不完美。

But safety isn’t the only selling point: Autonomous vehicles are said to cut traffic. But data from Waymo’s reports to the California Public Utilities Commission shows that, at least in that regard, robotaxis are no better than ride-hailing services like Lyft and Uber.

但安全性并不是唯一的卖点：人们常说自动驾驶汽车可以减少交通拥堵。然而，Waymo 向加州公用事业委员会提交的报告数据表明，至少在这一点上，自动驾驶出租车并不比 Lyft 和 Uber 等网约车服务表现更好。

Is there anyone in there? The study, published in Transport Findings by MIT Transit Lab Assistant Director of Research Awad Abdelhalim, analyzes data from August 2023 through December 2025, a roughly 1,000-day period. During that time, Waymo’s robotaxis completed 13.8 million trips for 19.3 million passengers over a total traveled distance of 86.3 million miles (138.8 million km), growing at a rate of around 15 percent a month.

车里有人吗？这项由麻省理工学院交通实验室助理研究主任 Awad Abdelhalim 发表在《Transport Findings》上的研究，分析了 2023 年 8 月至 2025 年 12 月期间（约 1000 天）的数据。在此期间，Waymo 的自动驾驶出租车为 1930 万名乘客完成了 1380 万次行程，总行驶里程达 8630 万英里（1.388 亿公里），并以每月约 15% 的速度增长。

Abdelhalim wanted to see what proportion of those rides were made by empty robotaxis—known as “deadheading”—and how the number changed over time. Initially, only 36 percent of Waymo’s miles were driven with a passenger onboard. But by the end of the study period, that had increased to around 56 percent and then plateaued, Abdelhalim found. So about 44 percent of Waymo’s driven miles are conducted with empty EVs.

Abdelhalim 想要了解这些行程中有多少比例是由空载的自动驾驶出租车完成的（即“空驶”），以及这一数字随时间的变化情况。研究发现，最初 Waymo 只有 36% 的行驶里程是有乘客在车上的。但到研究期末，这一比例增加到约 56% 后趋于平稳。这意味着 Waymo 约 44% 的行驶里程是在车辆空载的情况下完成的。

I’m not entirely surprised; on each of my recent visits to San Francisco, the sensor-festooned Jaguar I-Paces have been thick on the ground, but rarely did I spot any humans riding in them. In fact, there are two different kinds of deadheading: empty vehicles driving around waiting to be assigned a ride and empty vehicles driving to collect their passenger(s). And Waymo has been steadily reducing the number of miles driven empty en route to a pickup as it expands its fleet. The number of deadhead miles per trip has also been declining, in part due to Waymo’s introduction of freeway service, the author suggests.

我对此并不感到完全惊讶；在我最近几次访问旧金山时，到处都能看到装满传感器的捷豹 I-Pace，但我很少看到有人乘坐。事实上，空驶分为两种：一种是空车在路上徘徊等待派单，另一种是空车前往接载乘客。随着车队的扩大，Waymo 一直在稳步减少前往接客途中的空驶里程。作者指出，单次行程的空驶里程也在下降，部分原因是 Waymo 推出了高速公路服务。

A similar analysis conducted late last year on Waymo’s CPUC data from January 2024 through September 2025 by Matthew Raifman, who studies policy and autonomous vehicles at UC Berkeley, also found that 44 percent of Waymo’s miles were driven with empty vehicles and that two-thirds of those empty miles were robotaxis driving around waiting to be assigned a customer.

加州大学伯克利分校研究政策与自动驾驶汽车的 Matthew Raifman 在去年底对 Waymo 2024 年 1 月至 2025 年 9 月的 CPUC 数据进行了类似分析，同样发现 Waymo 44% 的里程是空车行驶，其中三分之二的空驶里程是自动驾驶出租车在路上徘徊等待派单。

No better, no worse than ride-hailing. Interestingly, similar arguments about reducing traffic were once made about ride-hailing. In 2014, other researchers at MIT published a study claiming that ride-hailing could reduce car ownership and cut traffic. Two of the authors later walked back their conclusions after evidence showed that ride-hailing actually increased traffic and CO2 emissions, partly because it was cheap enough to encourage trips people otherwise wouldn’t have taken. They noted that robotaxis would probably fall into the same trap. (A 2018 study found that almost half of the increase in vehicle miles traveled in San Francisco was attributable to ride-hailing services.)

与网约车相比，不分伯仲。有趣的是，关于减少交通拥堵的类似论点也曾出现在网约车身上。2014 年，麻省理工学院的其他研究人员发表了一项研究，声称网约车可以减少汽车拥有量并缓解交通压力。后来，其中两位作者收回了他们的结论，因为证据表明网约车实际上增加了交通流量和二氧化碳排放，部分原因是它足够便宜，鼓励了人们进行原本不会进行的出行。他们指出，自动驾驶出租车很可能会陷入同样的陷阱。（2018 年的一项研究发现，旧金山车辆行驶里程的增加中，近一半归因于网约车服务。）

In total, about 40 percent of the miles traveled by a Lyft or Uber driver are deadhead miles, suggesting there’s little difference in congestion whether there’s a human behind the wheel or not. Incidentally, this fact helps explain some of the statistical safety advantage of a robotaxi—if the average number of occupants of a robotaxi is always lower than the average number of occupants of a ride-hailing vehicle, the expected injury rate for the robotaxi should be correspondingly lower.

总的来说，Lyft 或 Uber 司机行驶的里程中约有 40% 是空驶里程，这表明无论是否有司机驾驶，对交通拥堵的影响几乎没有区别。顺便提一下，这一事实有助于解释自动驾驶出租车在统计学上的部分安全优势——如果自动驾驶出租车的平均载客量始终低于网约车，那么自动驾驶出租车的预期伤害率也应相应降低。

Meanwhile, effective congestion reduction could be achieved through a robust expansion of public transport. The same number of people on a bus takes up much less room on the road than if they were spread out in passenger cars, and the numbers get even better for trains and subways. But public transport doesn’t come cheap. Waymo might have raised $16 billion earlier this year for its robotaxis, and at least $100 billion has been invested in the sector since the 2010s. Meanwhile, the American Public Transport Association called for $268 billion in investment over five years, and a report by Transportation For America puts the price tag for a “world class” transit system at $4.6 trillion over the next 20 years.

与此同时，通过大力扩展公共交通可以实现有效的拥堵缓解。同样数量的人乘坐公交车在道路上占用的空间远小于分散在私家车中，而火车和地铁的效果则更好。但公共交通并不便宜。Waymo 今年早些时候可能为其自动驾驶出租车筹集了 160 亿美元，自 2010 年代以来，该行业已至少投入了 1000 亿美元。与此同时，美国公共交通协会呼吁在五年内投资 2680 亿美元，而“美国交通”（Transportation For America）的一份报告则指出，建立一个“世界级”交通系统的成本在未来 20 年内将达到 4.6 万亿美元。