ADAS是自动驾驶的必经之路，但还要等30年！

我想我是唯一对目前各家商业媒体争相报导自动驾驶车辆、积极预测其光明前景感到有点不安的人──确实，看着 Google Car 如何在加州Mountain View的街上穿梭自如是令人兴奋的，但当我听到 BMW 预测，完全自动驾驶车辆至少要到2025年才有机会问世，我忍不住想叹气：在这个因特网的时代，十年就已经是永无休止的等待了… 但想象一下，在2025年有个自动驾驶车辆的“驾驶人”，试图从俄罗斯开车到德国、或是从纽约开车到佛罗里达州？如市场研究机构IHS Technology的车用半导体市场首席分析师De Ambroggi所言，每个国家、或是美国的每一个州，对于车辆都有不同的法令规章，要在十年内调和所有不同规范恐怕不是简单任务。 我猜测，今日所讨论的自动驾驶车辆，将会分成不同阶段问世；每个阶段的自动驾驶车辆都有需要遵守的明文规定。 显然，汽车产业与科技供货商们都有充分动机去描绘自动驾驶车辆的光明未来；不只是BMW，几乎每一家车厂都热衷于重复宣传新车款上的多种先进自动驾驶辅助系统(advanced driving assistance system，ADAS)功能。而每家车厂都希望能首先获得新车安全评鉴协会(New Car Assessment Program，NCAP)──欧洲或美国──的整体性五星级评鉴。 但是当汽车厂商开始将ADAS做为迈向自动驾驶车辆时代的重要第一步，我们还不需要屏息以待，因为高度自动驾驶车辆与ADAS其实根本是不同的东西；实际上，两者之间有很大的差距。在解剖自动驾驶车辆之前，我们先来仔细看看ADAS市场。 车用半导体供货商都对于ADAS芯片的强劲成长数字兴高采烈，如IHS最近就预测，全球ADAS芯片市场到2020年可达26亿美元规模，是2010年时的 6.438亿美元之四倍以上。驱动ADAS系统的零组件将不只有传感器芯片、雷达与摄影机，不过这几种是组成今日ADAS芯片市场的主要部份。 其他对ADAS市场成长有贡献的零组件，还包括微处理器、微控制器、FPGA与内存；IHS的Ambroggi表示，如果说影像传感器等同于汽车的眼睛，其他这几种组件就是车子的大脑。而IHS预测，微处理器与内存将在2020年占据整体ADAS芯片市场的五成比例。 至于那些处理器组件将整合多少功能，以及执行多少软件(或应用程序)，端看系统架构以及每个处理单元的功能性。Ambroggi指出，车厂与芯片供货商选择 采用模块化布线是出自于相同的考虑，这种模式能更容易将安全关键硬件与非安全性硬件隔开，在调整系统时也更具弹性，端看每种不同车款所搭载之ADAS先进 程度。 无论如何，一辆汽车要配备多少先进的驾驶辅助功能，也决定了车辆对处理器、雷达与传感器的需求程度。 本文授权编译自EE Times，版权所有，谢绝转载 本文下一页：ADAS的技术演进

相关阅读：
• 飞思卡尔与东软、Green Hills Software携手推出面向ADAS视觉应用的开发生态系统
• 安森美半导体看好中国汽车市场
• 未来的出租车没司机、没方向盘并且自动驾驶npjesmc

{pagination} ADAS的技术演进 ADAS 实际上有两大类，被动式与主动式。举例来说，被动式ADAS在车辆偏离车道中心时会对驾驶人发出警告，主动式ADAS则会实际对车辆做出干预、藉由轻推方 向盘让车辆回到车道中央；而如果有必要，主动式ADAS在发现前方道路有立即性危险时会启动煞车，被动式ADAS在此时同样只是对驾驶人发出警告。 配备主动式ADAS功能如自动紧急煞车(automatic emergency braking，AEB)、自适应巡航控制(adaptive cruise control，ACC)以及车道维持辅助(lane keeping assist，LKA)等，我们得要求汽车辨识交通号志与路况、做出适当的决定并从驾驶人手中取得掌控方向盘权力。 当然，这种程度的控制会需要内建一套备援系统(redundant system)，因此额外的运算能力将可重复检查实际路况。有部分ADAS功能如自动煞车、自动转向以及跟车行驶(platooning)，预期可在接下 来五年问世，不过IHS的Ambroggi强调，这还是跟自动驾驶车辆不一样。 Ambroggi指出，为了要让车辆自动驾驶，车厂需要首度将所有系统链接；独立的ADAS功能不应该再独立运作，需要整合到一个整体系统中，车辆才能自己由地点A行驶到地点B，完全不需要驾驶人的干预；但就算到那个时候，所谓的自动驾驶车辆还是不等同于“全自动驾驶车辆(fully autonomous cars)”。 他预见自动驾驶车辆一开始将会受到不同的限制，包括驾驶人在何时、何地可使用自动驾驶模式，以及如何使用；更具体地说，将会有特定法规限制自动驾驶仅能在某种道路上执行，以及其行驶速率、可适用的天候状况(例如不可在下雪天执行)等等。 只要想一想你就会发现，其实会有很多需要限制自动驾驶的地方。而我的看法是，如果30年之后有人能够不受任何限制、大摇大摆地坐在一辆全自动驾驶车辆里，那将会是件很让人惊讶的事──想要一辆自动驾驶汽车吗？再等个30年看看吧！ 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：Fully Autonomous Car? Don't Buy Shotgun Yet，by Junko Yoshida

相关阅读：
• 飞思卡尔与东软、Green Hills Software携手推出面向ADAS视觉应用的开发生态系统
• 安森美半导体看好中国汽车市场
• 未来的出租车没司机、没方向盘并且自动驾驶npjesmc

{pagination} Fully Autonomous Car? Don't Buy Shotgun Yet Junko Yoshida, Chief International Correspondent I doubt that I’m alone in feeling a bit disturbed by the business media’s breathless coverage of the autonomous car and the industry’s aggressive forecasts for it. Sure, seeing a glimpse of how Google Car is driving on surface streets in Mountain View, Calif., is exciting. But when I hear BMW predicting fully automated cars by 2025, I can’t help but say, “Oh, come on, man.” Granted: Ten years is an eternity in the Internet era. But picture a “driver” in an autonomous car in 2025 trying to go from Russia to Germany, or New York to Florida? Each country and, in the case of the United States, each state, has different vehicle regulations. Aligning all the laws and regulations in 10 years will be no easy task, says Luca De Ambroggi, principal analyst for automotive semiconductors at IHS Technology. I suspect what’s discussed today as an "autonomous car" is likely to emerge in phases. These cars in each phase will come with fine print stipulating the conditions under which they must be driven. Obviously, the automotive industry and technology suppliers have every incentive to paint a bright future for autonomous cars. It’s not just BMW. Practically every carmaker is hot to trot out multiple ADAS (advanced driving assistance system) features in their new models. NCAP effect Everyone's gunning for the top overall rating of five stars from the New Car Assessment Program (NCAP) -- either the Euro or North American version -- which publishes safety reports on new cars. But when car companies start pitching ADAS as an important first step for the self-driving car era, well, let’s not hold our breath. After all, highly automated cars and autonomous cars aren’t the same thing. In fact, there’s a huge gap. Before dissecting autonomous cars, let's take a closer look at the ADAS market. We see automotive semiconductor suppliers elated with big growth numbers for ADAS chips. IHS, for one, recently predicted worldwide revenue for ADAS chips to “reach $2.6 billion in 2020, more than fourfold the size of the market of $643.8 million a decade earlier in 2010.” Driving ADAS won’t be just sensor ICs, radar, and cameras, although they are a big part of what constitutes the ADAS chip market today. Other components contributing to ADAS growth include microprocessors, microcontrollers, FPGAs, and memory. If image sensors are equivalent to “cars’ eyes,” these are the “cars’ brain,” says Ambroggi. IHS predicts that the MPU/memory segment will eventually hold 50 percent of the ADAS chip market by 2020. How much functional integration takes place in each of these processing units, and how much more software (or apps) each processor ends up running, will depend on the system architecture and the capability of each processing unit. There is an equal argument for car OEMs and chip vendors to opt for the modular route, Ambroggi told me. It allows easy decoupling between safety-critical hardware and non-safety-critical ones, and makes adjustment of a system more flexible, depending on the attach rate of advanced features for each model. Either way, how many more “advanced” driver assistance features a car is designed to offer determines the level of the processors, radars, and sensors the car requires. Evolution from ADAS to autonomous to fully autonomous ADAS, in fact, comes into two classes: passive and active. Passive ADAS, for example, warns a driver when the car starts to veer off the center of a lane. It takes active ADAS for the car to actually intervene, by nudging the steering wheel and putting the car back on center. Active ADAS, if necessary, would brake the car if it sees imminent danger on the road ahead. In contrast, passive ADAS would simply warn the driver. In implementing active ADAS functions such as automatic emergency braking (AEB), adaptive cruise control (ACC), and lane keeping assist (LKA), we are asking the car to interpret road signs and road conditions, make an appropriate decision, and take the wheel from the driver. Surely, this level of control would require a redundant system to be built in, so that the extra computational power would double-check the situation. Some ADAS functions such as automatic braking, steering, and platooning are expected to emerge in the next five years, according to the IHS analyst. But again, that’s not the same thing as autonomous cars, he cautions. In order to make a car self-driving, carmakers need to first connect all the systems together, says Ambroggi. Separate ADAS features should no longer operate independently. They need to be integrated into one coherent system so that a car can drive itself from point A to point B without driver involvement. But even then, autonomous cars aren’t the same as “fully autonomous cars,” he says. He foresees the autonomous car initially coming with different constraints -- in terms of when, where, and how a driver can use the car in autonomous mode. More specifically, there will be specific constraints such as which streets accommodate autonomous driving, at what speed, under what weather conditions (no snow, for example), etc. That’s a lot of restrictions, when you come to think of it. My forecast? I’ll be surprised if I’m riding shotgun in a fully autonomous car -- without any of those constraints -- 30 years from now.