立即打开
特斯拉“全自动驾驶”上线,听名字就不安全?

特斯拉“全自动驾驶”上线,听名字就不安全?

David Z. Morris 2020-11-09
如果司机过于相信辅助技术,他们开车时可能会变得漫不经心,让效果适得其反。

上个月,当特斯拉开始让一些客户测试其新添加到汽车上的新自动驾驶软件时,联邦监管机构便开始频繁出手。监管机构表示,他们将“密切监控这项新技术”,并“采取行动,保护公众免受不合理的安全风险的威胁”。

其中一个安全风险可能就来自于特斯拉的软件名称——“完全自动驾驶”。以目前的形式,该软件可以接管司机处理停车、行驶、高速公路上的巡航和遇到红灯时的刹车,但它不能让车辆“自动驾驶”。

按照行业标准,特斯拉的系统被认为是先进的驾驶辅助系统(ADAS),而不是自动驾驶技术。ADAS系统有望全面提高道路安全程度,但也有人担心,如果司机过于相信辅助技术,他们开车时可能会变得漫不经心,让效果适得其反。

美国汽车协会交通安全基金会(AAA Foundation for Traffic Safety)于今年9月发表的一项研究暗示,ADAS系统“高度自动化”的名称和营销手段会增加司机分心或失责驾驶的风险。在这项研究中,90名驾驶员被告知他们将测试驾驶一辆具有驾驶员辅助功能的车辆。

一半的司机被告知该系统名为“AutonoDrive”(自动驾驶),并提供了介绍信息,重点是该系统将如何减轻他们的工作量。另一半则被告知他们将使用一种名为“DriveAssist”(辅助驾驶)的系统,驾驶时必须对车辆进行监控。

系统的实际能力被准确地描述给每一组参与测试的驾驶员。

在此之后的调查中,“自动驾驶”组中有42%的司机高估了系统的能力,而“辅助驾驶”组中只有11%的司机犯了这一错误。“自动驾驶”用户还表示,在使用该系统时,他们“更愿意做可能分散注意力或危险的行为”,而且更相信该系统能够防止碰撞。事实上,根据研究人员的说法,该系统不太可能在这类情况下阻止碰撞的发生。

随后,两组司机都试驾了装有凯迪拉克超级巡航驾驶辅助系统的车辆(不过该系统的真实名字没有透露给司机)。被告知使用“自动驾驶”的司机把手离开方向盘的时间比被告知使用“辅助驾驶”的司机多30%。“自动驾驶”组的受试者对汽车自动控制系统发出的控制车辆的指令反应更慢,超过5秒。

美国汽车协会的汽车工程和工业关系主管格雷格•布兰农说:“这样的结果应该不会太令人惊讶。如果系统使用的名称拔高了系统的能力,无疑是在给司机设立了一个错误的情境:这个系统比它实际可以做到的还要厉害。”

布兰农并没有直接参与9月的研究,但他领导了一个由美国汽车协会和其他汽车工业组织发起的计划,为驾驶员辅助和自动驾驶系统设定标准化的命名规则。

像特斯拉这样的系统设计,其初衷是随着时间的推移而不断改进,但实际上可能反而导致用户愈发困惑。特斯拉及其首席执行官埃隆•马斯克曾经多次预测,完全的自动驾驶软件最终将使特斯拉能够在无人监督的情况下实现自动驾驶,甚至可以像机器人出租车一样运行。根据汽车工程师协会(Society For Automotive Engineers)的分类,这至少是四级自动化。但今天,特斯拉仍然强调,完全自动驾驶需要“驾驶员的主动监督”,意味着它仅符合SAE级别的二级自动化。

由此产生的混乱相当普遍:一些汽车出版物直言该软件为“完全自动驾驶”,而特斯拉明确否认。一些监管机构对这种脱节表示不满——例如,德国一家法院在7月裁定,名为“Autopilot”的商标具有误导性。

特斯拉并没有回复《财富》杂志就其命名做法置评的请求。

布兰农表示,特斯拉并不是美国汽车协会唯一担心的,沃尔沃的PilotAssist和戴姆勒的DrivePilot也可能使司机混淆。美国联邦贸易委员会(Federal Trade Commission)或将混淆名称纳入检查范畴,该委员会负责检查执行广告规则中的真实性。不过,美国汽车安全监管机构还没有制定任何命名标准,布兰农也不希望他们这么做。

布兰农说:“我们知道,汽车制造商可能会继续以不同的名称对这些系统进行个性化营销。”

相反,他希望制造商为特定的系统特性采用标准的名称,比如自动车道保持或自动紧急制动,并将其置于更广泛的系统品牌名下。这有助于减少驾驶员的混淆感,同时也通过统一测试使得比较不同车辆的性能更加容易。

但是,自动驾驶汽车行业的一些人更喜欢更为谨慎的术语。

初创公司Gatik的首席执行官瓜塔姆•纳朗表示:“作为一个行业,让公众对这项技术产生正确的认识非常重要。”该公司正与沃尔玛等公司合作,为行驶固定路线的货车、卡车开发自动配送系统。

纳朗说,这些基于可预测性而选择的固定路线,更适合真正的自动驾驶汽车现在的能力。他担心,承诺太多,像特斯拉这样的制造商可能会造成混乱——一旦出错,带来的将是长期的幻灭。

“标榜‘完全自动驾驶’是一个可怕的矛盾。”纳朗说,“因为如果没有密切的真人监督,它恐怕得变成‘完全无法驾驶’。”(财富中文网)

编译:杨二一

上个月,当特斯拉开始让一些客户测试其新添加到汽车上的新自动驾驶软件时,联邦监管机构便开始频繁出手。监管机构表示,他们将“密切监控这项新技术”,并“采取行动,保护公众免受不合理的安全风险的威胁”。

其中一个安全风险可能就来自于特斯拉的软件名称——“完全自动驾驶”。以目前的形式,该软件可以接管司机处理停车、行驶、高速公路上的巡航和遇到红灯时的刹车,但它不能让车辆“自动驾驶”。

按照行业标准,特斯拉的系统被认为是先进的驾驶辅助系统(ADAS),而不是自动驾驶技术。ADAS系统有望全面提高道路安全程度,但也有人担心,如果司机过于相信辅助技术,他们开车时可能会变得漫不经心,让效果适得其反。

美国汽车协会交通安全基金会(AAA Foundation for Traffic Safety)于今年9月发表的一项研究暗示,ADAS系统“高度自动化”的名称和营销手段会增加司机分心或失责驾驶的风险。在这项研究中,90名驾驶员被告知他们将测试驾驶一辆具有驾驶员辅助功能的车辆。

一半的司机被告知该系统名为“AutonoDrive”(自动驾驶),并提供了介绍信息,重点是该系统将如何减轻他们的工作量。另一半则被告知他们将使用一种名为“DriveAssist”(辅助驾驶)的系统,驾驶时必须对车辆进行监控。

系统的实际能力被准确地描述给每一组参与测试的驾驶员。

在此之后的调查中,“自动驾驶”组中有42%的司机高估了系统的能力,而“辅助驾驶”组中只有11%的司机犯了这一错误。“自动驾驶”用户还表示,在使用该系统时,他们“更愿意做可能分散注意力或危险的行为”,而且更相信该系统能够防止碰撞。事实上,根据研究人员的说法,该系统不太可能在这类情况下阻止碰撞的发生。

随后,两组司机都试驾了装有凯迪拉克超级巡航驾驶辅助系统的车辆(不过该系统的真实名字没有透露给司机)。被告知使用“自动驾驶”的司机把手离开方向盘的时间比被告知使用“辅助驾驶”的司机多30%。“自动驾驶”组的受试者对汽车自动控制系统发出的控制车辆的指令反应更慢,超过5秒。

美国汽车协会的汽车工程和工业关系主管格雷格•布兰农说:“这样的结果应该不会太令人惊讶。如果系统使用的名称拔高了系统的能力,无疑是在给司机设立了一个错误的情境:这个系统比它实际可以做到的还要厉害。”

布兰农并没有直接参与9月的研究,但他领导了一个由美国汽车协会和其他汽车工业组织发起的计划,为驾驶员辅助和自动驾驶系统设定标准化的命名规则。

像特斯拉这样的系统设计,其初衷是随着时间的推移而不断改进,但实际上可能反而导致用户愈发困惑。特斯拉及其首席执行官埃隆•马斯克曾经多次预测,完全的自动驾驶软件最终将使特斯拉能够在无人监督的情况下实现自动驾驶,甚至可以像机器人出租车一样运行。根据汽车工程师协会(Society For Automotive Engineers)的分类,这至少是四级自动化。但今天,特斯拉仍然强调,完全自动驾驶需要“驾驶员的主动监督”,意味着它仅符合SAE级别的二级自动化。

由此产生的混乱相当普遍:一些汽车出版物直言该软件为“完全自动驾驶”,而特斯拉明确否认。一些监管机构对这种脱节表示不满——例如,德国一家法院在7月裁定,名为“Autopilot”的商标具有误导性。

特斯拉并没有回复《财富》杂志就其命名做法置评的请求。

布兰农表示,特斯拉并不是美国汽车协会唯一担心的,沃尔沃的PilotAssist和戴姆勒的DrivePilot也可能使司机混淆。美国联邦贸易委员会(Federal Trade Commission)或将混淆名称纳入检查范畴,该委员会负责检查执行广告规则中的真实性。不过,美国汽车安全监管机构还没有制定任何命名标准,布兰农也不希望他们这么做。

布兰农说:“我们知道,汽车制造商可能会继续以不同的名称对这些系统进行个性化营销。”

相反,他希望制造商为特定的系统特性采用标准的名称,比如自动车道保持或自动紧急制动,并将其置于更广泛的系统品牌名下。这有助于减少驾驶员的混淆感,同时也通过统一测试使得比较不同车辆的性能更加容易。

但是,自动驾驶汽车行业的一些人更喜欢更为谨慎的术语。

初创公司Gatik的首席执行官瓜塔姆•纳朗表示:“作为一个行业,让公众对这项技术产生正确的认识非常重要。”该公司正与沃尔玛等公司合作,为行驶固定路线的货车、卡车开发自动配送系统。

纳朗说,这些基于可预测性而选择的固定路线,更适合真正的自动驾驶汽车现在的能力。他担心,承诺太多,像特斯拉这样的制造商可能会造成混乱——一旦出错,带来的将是长期的幻灭。

“标榜‘完全自动驾驶’是一个可怕的矛盾。”纳朗说,“因为如果没有密切的真人监督,它恐怕得变成‘完全无法驾驶’。”(财富中文网)

编译:杨二一

Federal regulators perked up when Tesla last month started letting some customers test new autonomous driving software that it had added to their cars. The regulators said they would “monitor the new technology closely” and “take action to protect the public against unreasonable risks to safety.”

One of those safety risks may be the name of Tesla's software: "Full Self Driving." In its current form, the software can take over from drivers to park, stay in a lane, cruise on highways and stop at red lights, but it does not make vehicles "self-driving."

By industry standards, Tesla's system is considered an advanced driver assistance system, or ADAS, and not autonomous driving technology. There is hope that ADAS systems will increase road safety overall. But there is also concern that if drivers place too much faith in assistive technology, they may become inattentive, undermining those safety benefits.

Names and marketing that imply higher levels of autonomy for an ADAS system increase the risk of distracted or irresponsible driving, according to a study published in September by the AAA Foundation for Traffic Safety. In the study, 90 drivers were told they were about to test drive a vehicle with driver assistance features.

Half of the drivers were told the system was called “AutonoDrive” and given introductory information that focused on how it would reduce their workload. The other half were told they would be using a system called “DriveAssist,” and that they had to monitor it.

The system’s actual capabilities were described accurately to each group.

When surveyed after this introduction, 42% of drivers in the “AutonoDrive” group overestimated the capabilities of the system, while only 11% of drivers in the “DriveAssist” group made the same mistake. “AutonoDrive” users also reported greater “willingness to engage in potentially distracting or risky behaviors” while using the system, and greater belief that the system would prevent a collision. In fact, the system was unlikely to prevent crashes in the scenarios presented, according to researchers.

Both groups of drivers then test-drove vehicles equipped with Cadillac’s SuperCruise driver assistance system (though the system's real name was not disclosed). Drivers who had been told they were using “AutonoDrive” spent 30% more time with their hands away from the wheel than those told they were using “DriveAssist.” Subjects in the “AutonoDrive” group were also much more likely to respond slowly (more than five seconds) to directions from the car's automated system to take control of the vehicle.

“The results shouldn’t be very surprising,” says Greg Brannon, AAA’s director of automotive engineering and industry relations. “If you use a name that overestimates the capabilities of the system … it sets a context for a driver that the system perhaps can do more than it can.”

Brannon was not directly involved in the September study, but he leads an initiative by AAA and other auto industry groups to set standardized naming conventions for driver-assistance and automated driving systems.

The fact that systems like Tesla’s are designed to improve over time could add to user confusion. Tesla and its CEO, Elon Musk, have frequently predicted that the Full Self Driving software will eventually enable a Tesla to drive itself without any human oversight, and even operate as a robotic taxi. That would be considered at least Level 4 automation, according to classifications by the Society For Automotive Engineers. But today, Tesla still emphasizes that Full Self-Driving requires “active driver supervision,” meaning it qualifies as Level 2 automation on the SAE scale.

The resulting confusion is widespread: even some automotive publications refer to the software as “fully autonomous” when Tesla explicitly says it’s not. Some regulators have cried foul at the disconnect – a German court, for instance, ruled in July that “Autopilot” branding was misleading.

Tesla did not reply to requests for comment from Fortune about its naming practices.

Brannon says Tesla isn’t AAA’s sole concern, citing names like Volvo’s PilotAssist and Daimler’s DrivePilot as also potentially confusing. Confusing names could become a matter for the Federal Trade Commission, which enforces truth in advertising rules. But U.S. auto safety regulators haven't created any naming standards, though, and Brannon doesn't expect them to.

“We understand that automakers are likely to continue the individualized marketing of these systems under different names,” Brannon said.

Instead, he hopes manufacturers adopt standard names for specific system features, such as automatic lanekeeping or automatic emergency braking, under the broader umbrella of their system brand names. That could help reduce driver confusion, while also making it easier to compare the performance of different vehicles' features using uniform tests.

Some within the autonomous vehicle industry, though, would prefer even more cautious terminology.

“As an industry, we are at the point where having the right public perception of the technology is important,” says Guatam Narang, CEO of Gatik, a startup that's working with companies like Walmart to develop autonomous delivery systems for vans or trucks that travel fixed routes.

Narang says those fixed routes, chosen for their predictability, are a better fit for what truly autonomous vehicles are capable of right now. He’s concerned that by promising too much, manufacturers like Tesla could create confusion – and perhaps, when things go wrong, long-term disillusionment.

“The branding 'Full Self Driving' is a terrifying contradiction,” says Narang, "because it can't actually be driven without close human supervision."

热读文章
热门视频
扫描二维码下载财富APP