无人驾驶大战，究竟谁占鳌头？

论起无人驾驶车领域的竞争，可以说既是暗战，也是公开的交锋。

汽车制造商，科技公司，还有创业公司经常通报目前的状况。但没有数据支持很难验证各公司的说法，也没法判断到底哪家公司的无人驾驶技术最先进。

加州汽车管理局综合了11家汽车和科技公司的资料后公布了一系列报告，向公众介绍了自动驾驶技术的发展状况。

如果想从加州汽车管理局申请测试许可，汽车公司得提交年度报告详细描述所谓的“脱离自动模式”情况，这个术语意思是测试中出现多少次因为软件问题或出于安全原因，驾驶员不得已恢复人工驾驶。例如，如果遇到其他车辆逆向行驶，或因行人、骑自行车的人违反交通规则而恢复人工驾驶。

加州汽车管理局周三发布的年报中提供了一些信息，从中能看出汽车制造商和相关企业在无人驾驶车大战中的状况。其中脱离自动模式次数并非唯一的指标。各家企业的无人驾驶车测试地点包括公开道路、研究中心和位于其他州的本公司场地。但对于确实提交报告的企业，确实能从中看出技术开发状况，从而了解发展动向。

Waymo（谷歌）是领头羊

随便扫一眼加州汽车管理局的报告就能看出，谷歌旗下的自动驾驶项目Waymo在加州公开道路上测试的总里程数遥遥领先。如今Waymo还单独成立了公司。

数字非常惊人，而且能看出该公司已经升级了自动驾驶软件。

2016年Waymo公司在公开道路的测试里程数近63.6万英里（约合102.4万公里），较之前一年增长了49%。而且，“脱离自动模式”从2015年的341次降到去年的124次，降幅达64%。

报告显示，总体来看，一年内每一千英里自动转人工操作的比例从0.80降至0.20。值得一提的是，Waymo选择的测试路段多为比较复杂的城市和郊区路面，而不是没有儿童和自行车，交通基础设施也更简单的高速路面。

“每次出现脱离自动驾驶事件后，我们都会借机改进完善汽车（我们设定的脱离自动驾驶标准非常保守），正因如此我们的技术进步很快，”Waymo公司自动驾驶技术负责人德米特里·多尔戈夫在媒体上一篇文章中写道。“对各种突发情况，我们可以即时创建上百种，有时甚至上千种相关场景，根据该区域其他道路参与者的不同位置和速度提供对应方案。”

通用汽车在加速赶超

从这份报告中也能看出大公司（以及相关资本和资源）如何推动创业公司成长。创业公司Cruise Automation是旧金山一家无人驾驶车技术开发商。2016年通用汽车出资超过10亿美元将其收购，此前该公司只能用一些日产Leaf电动车测试。

数字本身就能讲故事。2016年1月，Cruise只有三辆日产Leaf电动车做测试。4月测试车增加到五辆。一个月后，测试车代号为Scarlet和Quicksilver等的日产Leaf电动车逐渐减少，测试车变成了2017款雪佛兰Bolt EV电动车。

加州汽车管理局报告显示，8月时，日产Leaf已经完全不见，Cruise已经有至少九辆雪佛兰Bolt电动车，测试很频繁，代号类似叉角羚、山猫、鸭嘴兽和猎豹等。到了11月，该公司旗下已有20辆注册测试车。

特斯拉已将无人驾驶当成公司战略

2015年10月，特斯拉推出了半自动驾驶系统Autopilot，由此成为最先进的驾驶辅助软件开发商。Autopilot由感应器和软件组成，能协助完成路边停车、方向控制以及在高速路上变道。这款半自动驾驶系统一直在改进，领先其他制造商。

2016年之前，特斯拉一直没在加州公开道路测试过自动驾驶技术。2016年出现变化。汽车制造商和能源公司特斯拉提交加州汽车管理局的报告显示，2016年已开始在加州公开道路上测试四辆自动驾驶车。

有些汽车制造商基本没怎么在加州测试，或者测试很少

有21家公司，从大众到特斯拉到Zoox和Drive.ai等等初创公司申请过许可，在公开道路上测试无人驾驶技术。但只有11家公司提交了报告，其中还有一些报告内容只是声明没在公开道路上测试。

旗下包括豪车品牌奥迪的大众集团美国公司就没在加州公开道路测试。日本的本田也一样。本田表示目前在加州康特拉科斯达一处非公开的封闭道路上测试。宝马目前在美国仅有一台测试车。

其他在加州公开道路测试的企业包括：

• 博世：三台测试车

• 德尔福：两台测试车

• 福特：两台测试车

• 奔驰：一台测试车

• 日产：五台测试车

这些数字并不能代表某家企业实际的测试情况。很多制造商、科技公司和供应商都在其他州以及封闭道路上测试。举例来说，福特发言人透露，2016年其测试车队从10辆增加到30辆，而且到12月车队组建尚未完成。测试车队有多项工程目标，不只包括道路测试，而且测试地点包括亚利桑那州、加州和密歇根州。

不过要指出的是，过去加州一直是无人驾驶车开发中心。一些制造商在加州测试数量低可能因为正逐渐撤出加州，在其他更容易获得许可或是离制造商总部更近的州测试。

最大的问题是，其他没提交报告的公司都在忙什么？

报告内容非常不统一

报告并无统一标准，有些脱离自动驾驶的报告只有短短两页，有些像Waymo提交的则超过34页。

有些公司会追根溯源，提供脱离自动驾驶的具体地点和细节。有些则草草说明总里程，出现几次脱离自动驾驶，原因说明也只有一两段。

结果呢？由于报告形式各异，很难准确判断各家企业在开发无人驾驶技术方面的具体情况。

作者：Kisrten Korsec

译者：夏林

The race to develop autonomous vehicles can be described as a dual exercise in secrecy and bombastic theater.

Automakers, tech companies, and startups are often quick to share where they stand in that race. But without the data it's impossible to vet those claims or accurately evaluate which company has the most robust autonomous vehicle technology.

The California Department of Motor Vehicles released a series of reports Wednesday from 11 automakers and tech companies that gave the public a peek at their development of self-driving car technology.

Any company issued a testing permit from the California DMV must submit an annual report detailing its so-called "disengagements," a jargon term that means the number of times drivers have had to take control of a car because the software failed or for safety reasons. For example, a driver might take control if it encounters another car traveling the wrong way down the road or a pedestrian or cyclist acts erratically.

These annual reports, which the DMV released Wednesday, provide some insight into where automakers and tech companies stand in the race to develop autonomous vehicles. It's not the only measure, however. Companies are testing self-driving vehicle technology on public roads, research centers, and their own testing facilities other states. But for the companies that did submit reports, it does provide a hint of where they stand and what they're up to.

Waymo (Google) is leading

It takes only the quickest of glances at the reports filed with the California DMV to see that Waymo, the Google self-driving car project that spun out to become its own company, is way ahead in terms of total miles tested on public roads in California.

The numbers are striking—and show that the company has improved its self-driving software.

Waymo vehicles drove nearly 636,000 miles on public roads in 2016, a 49% increase from the previous year. And yet, the number of "disengagements" fell nearly 64% from 341 in 2015 to 124 last year.

In all, the total rate of disengagements per 1,000 miles driven fell from 0.80 to 0.20 over that one year time period, according to the report. It's also notable that Waymo tests in more complex urban and suburban environments, and not highways where there are fewer variables such as children, cyclists, and traffic infrastructure.

"We’ve been able to make dramatic improvements to our technology because we use each of these disengages to teach and refine our car (that’s why we set our thresholds for disengages conservatively), Dmitri Dolgov, head of self-driving tech at Waymo, wrote in a post on Medium. "For each event we can create hundreds — and sometimes thousands — of related scenarios in simulation, varying the parameters such as the position and speed of other road users in the area."

GM is quickly accelerating

This report is a study in how a big company (and capital and resources that come with it) can catapult a startup forward. Cruise Automation is a San Francisco-based developer of autonomous vehicle technology. And up until it was acquired by General Motors in March 2016 for more than $1 billion, the startup was testing its technology using a few Nissan Leaf vehicles.

The numbers tell the story. In January 2016, Cruise was testing three Nissan Leafs. That number topped out at five in April. A month later, the Nissan Leafs—with names like Scarlet and Quicksilver—began to drop off and 2017 Chevrolet Bolt EVs come onto the testing circuit.

By August, the Nissan Leafs were gone and Cruise had at least nine Chevy Bolts—with names like Pronghorn, Lynx, Platypus, and Cheetah—actively testing. By November, the company had 20 licensed vehicles in its testing lineup, according to the report filed with the California DMV.

Tesla has made it official

Tesla's semi-autonomous driving system Autopilot, which rolled out in October 2015, has made it a leader in advanced driver assistance. Autopilot, a combination of sensors and software, is capable of parallel parking, steering and lane changing on highways. The semi-autonomous system has continued to improve, pushing it ahead of other automakers.

And yet, until 2016 the company wasn't testing self-driving vehicle technology on public roads in California. That has changed. The automaker and energy company started testing four self-driving cars on California’s public roads in 2016, according to its disengagement report filed with the California DMV.

Some automakers didn't do any—or very little—testing in California

There are 21 companies—from Volkswagen and Tesla to startups like Zoox and Drive.ai—that have been issued permits to test self-driving vehicles on public roads. Only 11 filed reports, and some of those reports simply acknowledged that they did not test any vehicles on public roads.

Volkswagen Group of America, which includes luxury brand Audi, didn't do any public testing in California. Honda didn't either. Honda says it currently tests on a closed circuit, non-public course in Contra Costa County in California. BMW had just one vehicle testing in the U.S.

Other tallies of companies testing on public roads in California:

• Bosch: three autonomous vehicles

• Delphi: two autonomous vehicles

• Ford: two autonomous vehicles

• Mercedes: one autonomous vehicle

• Nissan: five autonomous vehicles

These low numbers don't necessarily reflect the level of testing by a particular company. Automakers, tech companies, and suppliers are testing in a variety of other states and on closed courses. ur overall For instance, Ford's overall fleet grew from 10 to 30 vehicles in 2016, although the build was not complete until December, according to a spokesperson. Those vehicles are used for a variety of engineering work, not just on-road testing and are spread throughout locations in Arizona, California, and Michigan.

It should be noted, that California has been focal point in the past for autonomous vehicle development. The low numbers could mean companies are pulling back from California and choosing to test in states with more permissive rules or closer to an automaker's base of operations.

The big question is what are the other 10 companies that didn't file reports up to?

These reports are wildly inconsistent

There is no standard, which means the disengagement reports vary from two pages to more than 34 from Waymo.

Some companies dig into the details, providing the location and specifics of a disengagement. Others simply state total mileage, the number of disengagements, and a brief one or two sentence explanation.

The upshot? It's difficult to provide an accurate assessment of where these companies are in the race to develop autonomous vehicle technology because of this variability.