五年后，Uber重返自动驾驶战场
五年前在美国亚利桑那州的街头，一辆正在测试的 Uber 自动驾驶汽车与一名行人相撞，悲剧瞬间发生。这一事故让一条生命戛然而止，美国国家运输安全委员会（NTSB）将 Uber 定为这个事故的部分责任方，这也让 Uber 的自动驾驶梦想按下了暂停键。
时隔五年，Uber 最近再次回到自动驾驶领域。上周，Uber宣布与电动车制造商 Lucid 以及自动驾驶技术公司 Nuro 合作，计划在未来六年里，在“全球数十个市场”部署约 2 万辆专为 Uber 平台打造的自动驾驶汽车。据说这批车将于明年开始生产，首批将在明年某座美国主要城市上路。为了进行这个合作，Uber 将向 Lucid 投资 3 亿美元，并向 Nuro 也投入数亿美元。
对于Uber的这个计划，很多业内人士却并非看好，觉得说这个计划是雄心勃勃算是轻的了，而这更像是一场“白日梦”。原因很简单，大规模自动驾驶最大的挑战仍是“可靠性”，特别是提供数据的传感器和硬件技术。
比如Nuro ，它的“Driver”系统配备了多种类型的摄像头、激光雷达、毫米波雷达和音频传感器；软件部分由多套 AI 模型组成，负责地图构建、感知、行为决策和车辆控制。虽然这些模型经过多年实地驾驶、封闭测试和模拟训练，但 Nuro 依然需要远程人工接管团队在无人驾驶车辆遇到状况时介入。
尽管系统设计已经十分先进，但潜在故障依旧存在：比如恶劣天气会干扰摄像头，雨水会削弱甚至瘫痪激光雷达，雷达也可能因环境干扰失灵。
我想，乘客愿意坐上机器人出租车，当然是希望它比人类司机更安全、更便利。但现实可能并不如愿。以 2023 年的数据为例，美国每行驶 1 亿英里约有 1.26 人死亡，全美车辆总行驶里程高达 3.19 万亿英里，也就是说，当年有超过 4 万人死于交通事故。这是一个建立在 2.84 亿辆汽车、数十亿次出行基础上的统计。
反观自动驾驶，Waymo 在洛杉矶、旧金山、菲尼克斯和奥斯汀的自动驾驶车队，累计行驶了 1 亿英里，没有发生重大事故。那听起来的确不错，但它的车队规模只有 1,500 辆，仅占全美车辆的 0.0005%。Nuro 在十多年间也只积累了大约 100 万英里的自动驾驶记录，同样没有“责任事故”，但样本量却过小，100 万英里在美国不过相当于 全国所有车辆 3 小时的总行驶量。
换句话说，这些“零事故”的漂亮成绩单的背后，是样本数量很小的行驶总里程，并不能直接推导出“自动驾驶更安全”的结论。要真正证明比人类驾驶更安全，需要的数据量至少要达到数十亿英里、覆盖各种复杂天气、光照、道路条件和突发事件。但目前，没有任何一家自动驾驶公司能接近这样的规模。
更何况，人类司机的事故率虽然不是零，但在绝大多数驾驶场景里，我们人类的判断和应变能力依然很强，尤其是在突发情况下。而自动驾驶系统一旦遇到训练数据之外的罕见场景，表现却往往难以预料。
对于Uber来说， 虽然有雄心大志，但是很多问题其实还并没有答案：比如安全责任由谁承担？远程接管员有多少？首批车会投放到哪些城市？会不会替代部分人类司机？车费如何定价？这些答案可能都要等到 2026 年才能揭晓。
那么问题那么多，为什么Uber要重新启动这新一轮的豪赌呢？
我觉得省钱固然是一个重要诱因，但背后真正的逻辑可能更为复杂。从长期来看，司机是 Uber 最大的运营成本之一，如今大部分车费都要分给司机，一旦有一天不再需要人工驾驶，那成本结构将被彻底改变，利润空间自然就更大。
但是更重要的还是这个行业的竞争。像Waymo这样的竞争对手已经在多个城市启动自动驾驶业务，如果 Uber一直依赖合作方提供技术，那么在未来的市场中就会不但缺乏主动权，还会被迫接受别人的规则。因此与其完全依赖外部，不如通过投资 Nuro 和 Lucid 来建立自己的技术与产品阵地。
另外，自动驾驶带来的不仅仅是运输本身，更是一台行驶中的数据库。每一辆车在道路上运行，都会采集大量交通、地理和驾驶行为数据，这些数据在人工智能训练、智慧交通规划、保险定价等各个领域都有巨大价值。
因此，我觉得Uber 这一次投资并不仅仅简单地是为了削减开支，而更是在为未来的市场版图和数据主导权提前布局。哪怕短期内无法替代人类司机，也要确保自己站在这场技术变革的中心位置。
那么你们觉得Uber 的这一轮新的投资是未来人类驾驶的转折点，还是另一场昙花一现的实验？还有你是否会去坐机器人出租车？欢迎大家给出你的答案。
Five Years After a Fatal Crash, Uber Bets Big on Robotaxis Again
Five years ago, on a street in Arizona, tragedy struck when a self-driving Uber test vehicle collided with a pedestrian. The crash claimed a life instantly, and the U.S. National Transportation Safety Board (NTSB) found Uber partially responsible. The incident forced the company to hit pause on its dream of a fleet of autonomous “robotaxis.”
Now, five years later, Uber is stepping back into the self-driving field. Recently, the company announced a partnership with electric vehicle maker Lucid and autonomous driving technology company Nuro. Over the next six years, they plan to deploy about 20,000 self-driving cars, purpose-built for Uber’s platform, across dozens of markets worldwide. The vehicles are expected to enter production next year, with the first batch hitting the streets of a major U.S. city in 2026. To make it happen, Uber will invest $300 million in Lucid and several hundred million dollars in Nuro.
Industry reaction has been far from uniformly optimistic. Many experts say calling the plan “ambitious” would be an understatement, with some dismissing it as a “daydream.” The main reason? Large-scale autonomous driving still faces its biggest challenge - reliability - especially when it comes to the sensors and hardware that feed critical data to the algorithms.
Take Nuro’s “Driver” system as an example. It’s equipped with multiple types of cameras, radar, and audio sensors, while the software stack contains several AI models for mapping, perception, behavior planning, and control. Despite years of real-world driving, closed-course testing, and simulations, Nuro still relies on remote human operators to step in when the autonomous vehicle runs into trouble.
Even with its advanced design, potential failures remain. For example, bad weather can distort camera input; rain can weaken or even disable lidar; radar can suffer from environmental interference.
Naturally, passengers who choose robotaxis expect them to be safer and more convenient than human-driven cars. But reality may fall short. In 2023, the United States recorded 1.26 fatalities per 100 million vehicle miles traveled. Total vehicle mileage reached a staggering 3.19 trillion miles, resulting in over 40,000 road deaths, a statistic based on 284 million vehicles and billions of trips.
By comparison, Waymo’s autonomous fleet in Los Angeles, San Francisco, Phoenix, and Austin has driven 100 million miles without a major accident. That sounds impressive, but the fleet has only 1,500 cars, just 0.0005% of the total U.S. vehicle count. Nuro, in over a decade of operation, has logged only about 1 million autonomous miles, also without an “at-fault” accident. Still, that mileage is tiny, roughly equal to all vehicles in the U.S. driving for just three hours.
In other words, these spotless “zero-accident” records are based on very small sample sizes. They don’t prove that autonomous driving is safer than human driving. To truly demonstrate superior safety, we’d need data covering billions of miles in all kinds of weather, lighting, road conditions, and unexpected events. Right now, no autonomous vehicle company comes close to that scale.
Moreover, while human drivers are not perfect, our judgment and ability to react in most driving situations, especially emergencies, remain strong. Autonomous systems, on the other hand, can behave unpredictably when confronted with rare scenarios outside their training data.
For Uber, many questions remain unanswered: Who will be responsible for safety oversight? How many remote operators will there be, and who will employ them? Which cities will get the first wave of vehicles? Will they replace some human drivers? How will fares be set? Most of these answers won’t be known until 2026.
So why is Uber making this big bet again, despite so many uncertainties? Saving money is certainly one motivation. Drivers are Uber’s biggest ongoing expense, with most of each fare going to them. If one day human drivers are no longer needed, the company’s cost structure would be fundamentally transformed and profit margins greatly improved.
But competition is an even bigger factor. Rivals like Waymo are already running autonomous services in multiple cities. If Uber relies entirely on partners for self-driving technology, it risks losing control over its future and being forced to play by others’ rules. Investing in Nuro and Lucid helps Uber establish its own technical and product foothold.
There’s also the data factor. A self-driving car is essentially a moving data center, collecting vast amounts of traffic, geographic, and driving behavior information. This data is invaluable for AI training, smart city planning, insurance pricing, and more. If Uber doesn’t control this layer, it risks being locked out of some of the most lucrative opportunities in future mobility.
In my view, this investment is less about short-term cost-cutting and more about positioning for control over the future market and the data that will drive it. Even if human drivers can’t be replaced in the near term, Uber wants to make sure it’s at the center of this technological shift.
So, what do you think? Is this bold new investment a turning point for human driving, or just another short-lived experiment? And would you ride in a robotaxi? I’d love to hear your thoughts.