无人驾驶和老司机，谁更安全吗？

未来20年，美国各大城市的面貌会变得很不一样，至少是在公共交通上。

这就是无人驾驶汽车公司Argo AI的CEO布莱恩•萨勒斯基的论断。今年二月，福特公司宣布将向这家创业公司投资10亿美元。自此，该公司便成了福特的下属企业。

本周三，萨勒斯基在加州半月湾举办的Rutberg FM科技大会上表示，无人驾驶汽车技术的兴起“去掉了驾驶中人的因素”，必将带来一种“安全得多的交通模式”。他认为，与未来的无人驾驶汽车相比，人类司机很容易分心和产生误判。

萨勒斯基曾任谷歌无人驾驶汽车项目的硬件主管。他认为，随着无人驾驶技术的普及，城市对停车位的需求也将越来越小。如今，各大城市的中心城区几乎有三分之一的面积都被用作停车场。而无人汽车将人送到目的地后，可以自动开走，不会占据停车位长达几个小时。这样一来，空下来的停车场就可以改造成写字楼或者公园。

萨勒斯基还预测道，购买了无人驾驶汽车的人在不使用汽车的情况下，还可以把自己的汽车变成无人驾驶出租车，让它自己出去拉客赚钱。他的这一想法也与特斯拉CEO伊隆•马斯克和百度前人工智能项目负责人吴恩达不谋而合，这两家公司也在研发无人驾驶技术。

福特最近刚刚用马克•菲尔德换掉了上任CEO吉姆•哈克特，该公司正希望依靠无人驾驶技术带动公司实现复兴。今年一月，马克•菲尔德曾表示，福特计划在2021年前开发一款完全自动驾驶的汽车。而萨勒斯基也再次确认，公司一定会赶在既定的日期完成任务。

不过萨勒斯基也承认，要想改进无人驾驶技术，福特还有大量的工作要做，尤其是在判断一些细微的差别时——比如当看到一个孩子独自在路上行走时，与他妈妈牵着他的手一起行走时，车辆的反应显然应该是不一样的。无人驾驶汽车应该知道，在没有父母监护的情况下，儿童的行为可能是“不规律”的，因此无人驾驶汽车“必须根据相应情况做出规划”。无人驾驶汽车要想成为一种主流技术，就必须具备感知大量路况细节的能力。

萨勒斯基表示，计算机技术的进步，使研究人员能够不断改进无人驾驶汽车的机器学习算法，从而使无人驾驶汽车能够在2021年前更好地“看懂”真实的世界。

萨勒斯基并没有直接拿福特的无人驾驶项目与谷歌的Waymo部门以及特斯拉和Uber等竞争对手的无人驾驶项目进行比较。不过他也提醒道，有些公司会鼓吹他们的无人驾驶汽车已经安全行驶了多少公里，外行人士对这种统计数据一定要留一个心眼。

比如特斯拉公司就在去年12月份表示，拥有部分无人驾驶功能的特斯拉汽车已经行驶了13亿英里。有些企业之所以乐于鼓吹这些里程数据，是因为它是“监管机构强制要求披露的少数指标之一”。

萨勒斯基表示：“里程和里程是不一样的，而且它也不是唯一的成功指标。”

他解释道，对于无人驾驶汽车来说，在高速公路上行驶的里程，并不像在城市环境中行驶的里程那样有价值，因为城市环境中有更多不可预测的变量——比如有行人突然横穿街道。而这种不可预测性有助于研究人员研发出更强大的无人驾驶技术，它将能够应对所有路面情况，而绝非仅仅是在路况相对单纯的高速公路上，毕竟高速公路是一个“对司机比较友好的环境”。（财富中文网）

译者：朴成奎

U.S. cities will look a lot different in 20 years, at least when it comes to public transportation.

That’s according to Bryan Salesky, the CEO of the self-driving car company Argo AI, which became a Ford Motor subsidiary after the auto giant said in February it would invest $1 billion in the startup.

The rise of self-driving cars will usher a “much safer mode of transportation” by “removing the human from the loop,” Salesky said on Wednesday at the Rutberg FM technology conference in Half Moon Bay, Calif. Human drivers are more prone to distractions and errors in their judgment compared to autonomous cars in the future, Salesky believes.

Salesky, who was once Google’s (goog, -0.28%) director of hardware for its self-driving car project, also believes that cities will have less need for parking spaces, which he said currently account for one-third of the space in an average urban city center. Self-driving cars would be constantly shipping people around cities instead of remaining parked for hours, and the unused parking spaces could be converted into office buildings or public parks, he explained.

People who buy self-driving cars will also have the option to convert their automobiles into autonomous taxis when they don’t use them, Salesky predicted. His thoughts echo recent statements made by Tesla (tsla, +1.19%) CEO Elon Musk and Andrew Ng, the former head of AI for Chinese search giant Baidu, which is also working on self-driving cars.

Ford, which recently replaced CEO Mark Fields with Jim Hackett, is hoping new technologies like self-driving cars can help revive the struggling company. Fields said in January that Ford planned to develop a completely autonomous vehicle by 2021, and Salesky confirmed that the auto company is sticking with that date.

But Ford has a lot of work to do to improve autonomous vehicle technology, Salesky admitted, especially in terms of creating autonomous cars that can perceive nuances like the differences between children by themselves and when they are holding their mother's hands. A self-driving car needs to know that children, when not by the side of their parents, may move "erratically," and so the car must "then plan accordingly." Self driving cars will need this ability to perceive tremendous amounts of detail on the road if they are to become mainstream.

Salesky says that advances in computing technology will make it possible for researchers to improve the machine learning algorithms powering self-driving cars, thus helping those cars better “see” the physical world by 2021.

The CEO did not directly compare Ford’s efforts on self-driving cars with competitors like Google’s Waymo unit, Tesla, or ride-sharing company Uber. But, he warned outsiders to be wary when these companies share statistics about how many miles their self-driving cars have driven.

Tesla, for example, said in December that Tesla cars with some self-driving capabilities have driven about 1.3 billion miles. Businesses brag about the number of miles their self-driving cars log because it’s “one of the few metrics the companies are forced to release” by regulators, he said.

“Not all miles are the same, and that’s not the only indicator of success,” Salesky said.

He explained that miles logged by self-driving cars on the highway are not as valuable to researchers as miles driven in urban environments, because city driving involves more unpredictable variables like people suddenly crossing streets. This unpredictability can be used to help researchers create more powerful self-driving cars that can handle all driving situations, not just highway roads, which are "a much friendlier environment."