当自动驾驶车辆出现异常…该怎么办？

在日前于日本东京举行的年度智能运输系统世界大会(ITS World Congress)上，几乎所有的目光焦点都聚焦于自动驾驶车辆，来自全球各地的汽车制造商、政府主管机关、技术供货商齐聚一堂，讨论议题并未多着墨于花俏的自动驾驶车辆本身，而是包括人机接口、安全性、可靠性，社会对自动驾驶车辆的接受度，以及整体所需新法规的订定等等。

opQesmc

不过在一场题目为“自动驾驶车辆实现之道(Autonomous Vehicles -- the Path to Implementation)”的座谈会上，与会专家巧妙地回避了一个很重要的议题：当自动驾驶车辆遭遇了计算机程序无法处理的危险，该如何将控制权交回给驾驶人？ 这应该是最近常关注EETimes网站一系列自动驾驶车辆报导的读者们都很想知道的；在先前的一篇读者留言(编按：EETimes美国版网站)中，一位读者以飞机的自动驾驶与自动驾驶车辆来比较，分享了他担任飞机驾驶员的兄弟的经验，写道： “飞机机师必须定期通过模拟机与教官的独立驾驶测试，以不同临场状况做为训练，例如观察他们在零失误(no error )飞行中的注意力，在一段高压力航程中的疲劳程度；或是观察恶劣天候飞行、艰困降落情况下飞行员的反应时间与控制精准度。以上这些都必须被评估。” 而像是 Google Car这样的自动驾驶车辆将如何处理一些“例外状况”，汽车厂商又预期驾驶人将如何在一些特定情况下收回方向盘控制权？EETime美国版有另外一位读者的看法是： “可 以确定的是，无论是汽车驾驶人或是飞机机师都需要准备好应付例外状况，那些状况可能是计算机系统永远无法处理的。若透过计算机操作是不方便的，对于驾驶车辆来 说就可能是致命危机──试想当车辆在时速60英里 (约96公里)的行驶状态，若计算机出现当机，恐怕不太可能出现什么好结局。” 本文授权编译自EE Times，版权所有，谢绝转载 第2页：飞机自动驾驶跟自动驾驶车辆不能相比？ 第3页：车子看见前方物体，越过驾驶人踩煞车，还是让驾驶人自己抉择？

相关阅读：
• 苹果预计明年将推出"车载iOS系统"
• Google Car总监：别妄想100%有效的汽车安全技术！
• 汽车电子缺陷酿车祸？丰田在美惹官司opQesmc

{pagination} Google 的自动驾驶车辆安全技术总监Ron Medford 在会后接受EETimes美国版编辑访问时，针对自动驾驶车辆如何处理例外状况的问题有些答非所问：“对于驾驶人如何在片刻之间接手自动驾驶车辆控制权的 问题，产业界还未深入了解。”他指出，目前有一些相关研究正在进行中，不只是Google等厂商，包括政府主管机关也很想知道研究结果。 飞机自动驾驶跟自动驾驶车辆不能相比？ Medford 原本是美国国家高速公路交通安全局(National Highway Traffic Safety Administration，NHTSA)的副局长，在今年初转任Google；他先前曾任职于美国消费者产品安全委员会(Consumer Product Safety Commission)，2003年进入NHTSA，2009坐上该单位的第二把交椅。 对于自动驾驶车辆与驾驶人对车辆控制权交手的议题，Medford显然曾经思考过，但他仅小心翼翼地表示，Google期望在某一天，该公司的自动驾驶车辆技术足以先进到能做出与人类相同的判断。但他也坦承，那将会是很远的未来。 面对同样的疑问，汽车电子供应大厂Continental Automotive日本分公司总裁暨首席执行官Christoph Hagedorn的回答则是：“你怎么能将利用自动驾驶系统开飞机的机师，跟自动驾驶车辆驾驶人画上等号？” Hagedorn 目前也是Continental汽车底盘与安全系统(Chassis and Safety)部门董事会成员，该业务总部位于德国，是关键的汽车技术供货商；他停顿了几秒又表示：“也许你可以将两种技术画上等号──飞机的自动驾驶与 自动驾驶车辆的自动驾驶模式，但两种交通工具的驾驶(飞行员与汽车驾驶人)则不行。” 换句话说：“你不能预期汽车驾驶人为了自动驾驶车辆接受额外训练；”Hagedorn表示：“我认为自动驾驶车辆的终极目标是为驾驶人提供‘舒适’。”若自动驾驶车辆的驾驶人被要求接受特殊训练、并必须在方向盘后时时保持警觉，那就失去了其意义。 Hagedorn指出，德国的法令规定，汽车与驾驶人控制权的交接时间必须在10秒之内，10秒也许是足够的，但更实际的是打造出在特定状况下──例如将驶离高速公路──能自动提醒车辆关闭自动驾驶模式的基础设施。 本文授权编译自EE Times，版权所有，谢绝转载 第3页：车子看见前方物体，越过驾驶人踩煞车，还是让驾驶人自己抉择？

相关阅读：
• 苹果预计明年将推出"车载iOS系统"
• Google Car总监：别妄想100%有效的汽车安全技术！
• 汽车电子缺陷酿车祸？丰田在美惹官司opQesmc

{pagination} 嵌入式视觉联盟(Embedded Vision Alliance)的创始人Jeff Bier最近接受EETims美国版访问时表示，光是一辆配备先进驾驶人辅助系统(ADAS)的汽车，行驶在一般道路上可能遭遇的挑战就很多，更别说是自动驾驶车辆。 他在一年前从售后市场买了一套ADAS系统，在高速公路上行驶非常稳，但一下高速公路，车子只要在不同的天候下遇到随机物体──电线杆、行人、路标──就会发出刺耳的误判警报。ADAS就可能面临无数种状况需要处理，自动驾驶车辆所遭遇的状况恐怕更复杂。 Continental的Hagedorn坦承，“故障安全(Fail safe)”这个名词，对自动驾驶车辆的意义大于字面解释。汽车制造商必须对系统重复进行测试，并且在公开道路上做广泛的评估。 如日本汽车研究所智能运输系统中心研究总监Keiji Aoki所言，除非自动驾驶车辆的可靠性议题能获得解决，其他非技术性议题都不会有结果；后者包括包括驾驶人与自动化系统所负担的责任，对自动驾驶车辆驾驶人义务的定义，以及其他相关法令规章。 对 打算为车辆提供ADAS的汽车制造商来说，如何针对做出煞车、转向决策的理想瞬间进行妥善设计，是一个典型的议题；这对自动驾驶车辆来说也是一大关键。举例来说，当一辆Volvo看见前方物体，是该主动越过驾驶人踩煞车，还是该让驾驶人自己抉择？Hagedorn表示，这种精确的分工通常都是由车厂自己决 定。 到底什么是实现自动驾驶车辆的最大问题？Hagedorn指出，最艰困的挑战会是让该类车辆通过主管机关核准：“不同的驾驶状况可能有上百万种，这需要花上几年的时间进行验证，该如何让符合安全标准的车辆获得批准？产业界需要针对此议题做出因应。” 最后，也许自动驾驶车辆的驾驶人还是得好好思考一个重要问题：“要是你的车在路上发“神经”，你准备好要怎么处理了吗？” 本文授权编译自EE Times，版权所有，谢绝转载 编译：Judith Cheng 参考英文原文：Are You Ready When Your Google Car Freaks Out?，by Junko Yoshida

相关阅读：
• 苹果预计明年将推出"车载iOS系统"
• Google Car总监：别妄想100%有效的汽车安全技术！
• 汽车电子缺陷酿车祸？丰田在美惹官司opQesmc

{pagination} Are You Ready When Your Google Car Freaks Out? Junko Yoshida TOKYO — All eyes have been fixed on the autonomous car at the ITS World Congress here this week. For carmakers, regulators, and technology suppliers who have gathered here from all over the world, the big concern is not so much the bells and whistles of self-driving cars, but more about deployment strategies for them. Debates here covered everything from human-machine interface, safety, and reliability to societal acceptance and a whole new legal landscape. One issue, however, was artfully dodged by panelists on a Tuesday, Oct. 15, executive session called "Autonomous Vehicles -- the Path to Implementation." Put simply, how does a self-driving car, when faced with a crisis beyond its program, hand control back over to the (human) driver? This issue is near and dear to the hearts of EE Times readers, as shown by the reader discussion at What’s Going Through This Driver’s Head? The debate was ignited when an EE Times reader compared a pilot flying a plane on autopilot and a driver in a self-driving car. Referencing his brother, a pilot for a major airline, the reader wrote: Pilots must be tested in simulators and by instructors riding along periodically, make these scenarios part of that training. Give them a no error flight and watch the attention span. Give them a stressful flight and watch for fatigue. Give them a bad weather flight with a tough landing; watch the reaction times and the precision of control. All these factors must be evaluated. At issue here is how well a self-driving car like Google Car can handle "exceptions" and how carmakers expect a driver to take over the control of the self-driving car in a critical situation. Another EE Times reader pointed out: This confirms the idea that drivers and pilots need to be prepared to handle the exception situations, which are exactly the type of things that computer systems will never be able to handle. It is inconvenient when using a computer, it could be fatal driving a car. Picture that "blue screen" at 60MPH, and you can realize that there would not be any good possible ending available. Asked by EE Times, after the ITS session, about how self-driving cars are designed to handle exceptions, Ron Medford, Google's director of safety for self-driving cars, deflected the question. He said, "Readiness of drivers to take over the control [of a self-driving car at the moment's notice] is simply not understood yet." Citing a number of studies now underway, Medford explained that, not only companies like Google, but also a lot of government officials and regulators are all eager for the results. Autopilot and the self-driving car: not the same thing? Medford, former deputy director of the National Highway Traffic Safety Administration, took a position at Google earlier this year. He held down the No. 2 spot at NHTSA since 2009, originally joining the organization in 2003 after a career at the Consumer Product Safety Commission. Medford, clearly, had thought about the issue. But he was careful in stating only that Google hopes for a day when Google's autonomous cars are advanced enough to take the human judgment out of the equation. But, as he acknowledged, that is still in far future. Continental's Christoph Hagedorn with a microphone (left) and Google's Ron Medford at ITS executive session entitled "Autonomous Vehicles -- the Path to Implementation." EE Times posed the same question to Christoph Hagedorn, president and CEO of Continental Automotive Corporation Japan: "Can you draw a parallel between a pilot flying an airplane on autopilot and a driver in a self-driving car?" Hagedorn, who also serves as a Divisional Board Member, Chassis and Safety, at Continental, one of the key automotive technology suppliers based in Germany, paused a few seconds and responded: "I think you can draw a parallel on the technology -- airplane on autopilot and autonomous car in self-driving mode, but not on the driver" (airplane pilot vs. consumer driver). In other words, "You can't expect drivers to do extra training for driving autonomous cars," Hagedorn said. "I think self-driving cars are ultimately about drivers achieving 'comfort.' " Drivers being asked to take more driver education courses and stay alert all the time behind the wheel inside a self-driving car, in a way, defeats the whole purpose of autonomous cars. Hagedorn noted that in Germany, rules require 10 seconds to hand over control from car to driver. A 10-second transition might seem adequate, but more realistic is the development of an infrastructure that automatically cautions and turns off the self-driving mode when a car is, say, getting off a highway, Hagedorn explained. Driving a car with an advanced driver assistance system (ADAS), let alone a self-driving car, on a regular surface road poses the greatest challenge. Jeff Bier, founder of the Embedded Vision Alliance, recently told EE Times that an after-market ADAS system he bought a year ago "runs rock solid on highways." But off the highway, his car sees random objects -- utility poles, pedestrians, and road signs -- under a variety of weather conditions and "tends to sound off shrill false alarms... There is just an infinite variety of conditions ADAS needs to deal with." With the self-driving car, more and worse complications are likely to crop up. "Fail safe" is a loaded word for self-driving cars, acknowledged Continental's Hagedorn. Carmakers must test their systems again and again, and do extensive evaluation on public roads. As Keiji Aoki, research director of the ITS Center at the Japan Automotive Research Institute, noted during the session that unless reliability issues are settled, other non-technical issues will not get resolved. Such thorny issues include the liability of the driver and automated systems, a definition of the human driver's obligations, and other regulatory worm-cans. How best to design the ideal moment for braking and steering decisions are typical issues for carmakers applying ADAS to their cars. The question of how to define and where to set the point of brake and steer is also critical to autonomous cars. Will a Volvo, seeing an object, override the driver and hit the brakes, or leave it up to driver's choice? Such refinement, said Hagedorn is "usually left up to a decision by individual carmakers." During the session, asked about problems keeping autonomous cars out of the real world, Hagedorn said the toughest challenge is the approval of a vehicle. "When there are millions of different driving scenarios, which will take years of validation, how do you approve a vehicle that's safe to put on a road? The industry needs to plan for that." But in the end, it is left up to an individual driver to ask himself the following question: "Will you be ready when your Google car freaks out on the road?"