柯蒂斯·莫爾德里奇 維多利亞·伍拉斯頓

Driverless cars used to be the sort of thing youd see in sci-fi1 films—but theyre becoming a reality. Autonomous car technology is already being developed by the likes of Lexus， BMW2 and Mercedes3， and weve even tested Teslas4 driverless Autopilot system on UK roads. Across the Atlantic， Google is developing its automated technology in the wild， and Apple is rumoured to be working with BMW on its own—probably automated—car.

Fully-driverless tech is still at an advanced testing stage， but partially automated technology has been around for the last few years. Executive5 saloons like the BMW 7 Series feature automated parking， and can even be controlled remotely.

With so much investment and interest in driverless technology， its easy to assume that self-operating cars are imminent， but before our roads are flooded with driverless vehicles， manufacturers must tackle a range of technical and ethical challenges， and combat the biggest threat to autonomous technology： humans.

The Google Car6

Autonomous vehicles rely on a range of sensors to interact with the world around them， with the Google Car prototype coming equipped with eight.

The most noticeable is the rotating roof-top LIDAR—a camera that uses an array of either 32 or 64 lasers to measure the distance between objects， building up a 3D map at a range of 200m and allowing the car to “see” hazards. The car also sports another set of “eyes”， a standard camera that points through the windscreen. This looks for nearby hazards like pedestrians， cyclists and other motorists7， as well as reading road signs and detecting traffic lights. Speaking of other motorists， bumper-mounted radar， already used in intelligent cruise control， tracks other vehicles in front of and behind the car.

Externally， the car has a rear-mounted aerial that receives geolocation information from GPS satellites， and an ultrasonic sensor on one of the rear wheels monitors the cars movements. Internally， the car has altimeters， gyroscopes and a tachometer （a rev-counter） to give even finer measurements on the cars position， all of which combine to give it the highly accurate data needed to operate safely.

Using these arrays8， the Google Car can read the road like a human， but these sensors come with their own limitations. Autonomous cars simply replace the human eye with a camera， leaving them vulnerable to extreme sunlight， weather or even defective traffic lights. In current autonomous cars， the way this selection of pixels is analysed could be the difference between a safe journey and death.

Connected cars

Many believe a connection between cars and traffic infrastructure is needed to combat this problem. “Car-to-car and car-to-infrastructure communication is essential for enabling autonomous driving，” says Christoph Reifenrath， senior manager in technology marketing of Harmans9 infotainment division， who supply in-car tech to the likes of Audi， BMW and Mercedes.

“For example， as your car approaches a red light， well give you information. How can we provide this information in every car at every red light？ There has to be a solution for that for if you want to enable autonomous driving in areas with traffic lights.”

The German automotive industry is one of the most powerful advocates of a connected car-traffic infrastructure. Earlier in 2018， manufacturers including Daimler， BMW and Audi paid $3.1 billion for the Nokia Here mapping service， which will be used as a platform for a connected-car environment. A joint statement released by the consortium reads：

To become a viable solution， these systems will be required in every vehicle， including those still used by humans. Its likely that emergency vehicles10 like ambulances and police cars will continue to use human drivers， so theyll need a method of communicating with the autonomous cars around them.

“You have to know where [an emergency vehicle] comes from and when it will be there， so the information is shared between this car and your car，” adds Reifenrath.

The human problem

Humans present problems for autonomous cars as both drivers and pedestrians， and dealing with our unpredictable behaviour represents a significant challenge for the technology.

The Google Car is one of the most experienced autonomous vehicles， and its interaction with human drivers has exposed one of driverless cars main weaknesses. The first injury involving the Google Car wasnt due to a fault in its system， but human-error.

While correctly waiting at traffic lights， Googles self-driving car was hit by an inattentive driver and， despite its sophisticated array of sensors， there was little it could do to avoid the incident. Luckily， the accident only resulted in whiplash11 for a few of the passengers， but its a reminder that autonomous cars are at risk when surrounded by human road users12.

Despite their sophisticated systems， self-driving cars currently have no plan B13 for human road users. Human drivers are able to interact with each other and make allowances14， but also make countless， small mistakes when driving—mistakes to which current self-driving cars simply cant adapt.

Dealing with pedestrians

The way human drivers interact with pedestrians raises difficult moral and ethical questions for car manufacturers—with implications.

Autonomous cars need to understand the way pedestrians behave， while also mimicking the behaviour theyd expect from a human driver. “Everyone has an appreciation15 of how a human being is going to react， because we are all human beings，” says computer ethics commentator Ben Byford. “So if you walk out in front of a car， and presumably the car driver knows youre there， theyre going to react in a certain way.”

“If I walked out in front of a Google Car travelling at 60mph， I have no real appreciation of how the vehicle will behave， so Im effectively putting myself at a disadvantage.”

過去，人們只會(huì)在科幻電影中見到無人駕駛汽車——但如今它們正逐漸走向現(xiàn)實(shí)。眼下，雷克薩斯、寶馬和梅賽德斯等汽車制造商都在研發(fā)無人駕駛技術(shù);特拉斯Autopilot自動(dòng)輔助駕駛系統(tǒng)甚至已在英國公路上進(jìn)行了上路測(cè)試。大西洋對(duì)岸，谷歌公司正在野外研發(fā)這項(xiàng)技術(shù)，據(jù)稱蘋果公司也在與寶馬合作開發(fā)自有品牌汽車（很可能是無人駕駛汽車）。

全自動(dòng)無人駕駛技術(shù)目前仍處于測(cè)試階段后期，但近幾年來半自動(dòng)駕駛技術(shù)已經(jīng)發(fā)展到了一定程度。諸如寶馬7系的豪華三廂轎車以自動(dòng)泊車為賣點(diǎn)，甚至具備遠(yuǎn)程控制功能。

在人們的大力投資和廣泛關(guān)注下，不難想象，無人駕駛汽車正離我們的生活越來越近。但在無人駕駛汽車主宰道路前，汽車制造商們必須解決一系列技術(shù)和道德難題，并設(shè)法應(yīng)對(duì)無人駕駛技術(shù)最大的威脅：人類。

谷歌無人駕駛汽車

無人駕駛汽車依靠各類傳感器感知周遭環(huán)境，與外部世界互動(dòng)。以谷歌無人駕駛汽車為例，其原型車配備了八大傳感器。

最引人注目的是安裝在車頂?shù)男D(zhuǎn)光學(xué)雷達(dá)LIDAR——一款使用32線或64線激光雷達(dá)測(cè)量物體間距離的攝像頭，它能在200米測(cè)距范圍內(nèi)構(gòu)建三維空間地圖，幫助車輛發(fā)現(xiàn)沿途的安全隱憂。鑲嵌在擋風(fēng)玻璃上的標(biāo)準(zhǔn)規(guī)格攝像頭是該車的另一雙“眼睛”，它不但能捕捉道路附近的各類安全威脅（如行人、自行車和其他車輛），還能識(shí)別路標(biāo)和信號(hào)燈。談到其他車輛構(gòu)成的安全威脅，值得一提的還有安裝在汽車保險(xiǎn)杠上的雷達(dá)，它能追蹤前后方車輛運(yùn)行狀況，已應(yīng)用于智能巡航控制系統(tǒng)。

在汽車外部，谷歌汽車的車尾處設(shè)有一條天線，可接收來自GPS衛(wèi)星的定位信息;此外，它的一只后輪上安裝了監(jiān)控車輛運(yùn)動(dòng)狀況的超聲波傳感器。汽車內(nèi)部則配備了高度儀、陀螺儀和測(cè)速儀（一種發(fā)動(dòng)機(jī)轉(zhuǎn)速表），用來更準(zhǔn)確地測(cè)量汽車定位，為汽車安全運(yùn)行提供精確的數(shù)據(jù)。

有了以上配置，谷歌汽車能像人類一樣讀懂路況，但這些傳感器難免存在缺陷：無人駕駛汽車僅僅是用攝像頭代替了人眼，這使其易受極強(qiáng)日光、天氣或交通信號(hào)燈故障等狀況的影響。然而，對(duì)目前的無人駕駛汽車來說，系統(tǒng)對(duì)攝像頭像素?cái)?shù)據(jù)的分析處理直接影響出行安全，事關(guān)生死。

互聯(lián)汽車

為應(yīng)對(duì)該問題，許多人認(rèn)為有必要將汽車和交通基礎(chǔ)設(shè)施互聯(lián)互通?！败噷?duì)車和車對(duì)基礎(chǔ)設(shè)施通信技術(shù)是實(shí)現(xiàn)自動(dòng)駕駛的必要條件?！笨死锼雇蟹颉べ嚪依卣f，賴芬拉特是哈曼國際工業(yè)公司信息娛樂部門技術(shù)營銷高級(jí)經(jīng)理，為奧迪、寶馬和梅賽德斯等汽車制造商提供車載科技支持。

“例如，我們會(huì)在汽車靠近紅燈時(shí)給予信息提示。如何讓每輛汽車在遇到紅燈時(shí)都獲得提示？若想在有交通信號(hào)燈的區(qū)域?qū)崿F(xiàn)無人駕駛，必須提供該問題的解決方案?！?/p>

德國汽車工業(yè)是車與交通基礎(chǔ)設(shè)施互聯(lián)互通建設(shè)最具影響力的支持者之一。2018年年初，包括戴姆勒、寶馬和奧迪在內(nèi)的汽車制造商以31億美元收購了諾基亞Here地圖服務(wù)，將它用作互聯(lián)汽車技術(shù)支持平臺(tái)。這些汽車制造商聯(lián)合發(fā)布了如下聲明：

欲使該方案切實(shí)可行，需要給所有汽車都配備這些系統(tǒng)，包括人工駕駛的汽車。因?yàn)榫茸o(hù)車和警車這類執(zhí)行緊急任務(wù)的特種車輛可能仍由駕駛員駕駛，應(yīng)想辦法讓它們與周圍的無人駕駛汽車建立通信。

“無人駕駛汽車需要知曉特種車輛會(huì)在何時(shí)從何處駛來，因此兩輛車之間得共享信息。”賴芬拉特補(bǔ)充道。

涉及“人”的難題

無論是作為駕駛員還是行人，人類都給無人駕駛技術(shù)帶來了不少難題，應(yīng)對(duì)人類無法預(yù)料的行為是無人駕駛技術(shù)面臨的一項(xiàng)重大挑戰(zhàn)。

對(duì)谷歌汽車來說，無人駕駛可謂最駕輕就熟的了，但它在與人類駕駛員互動(dòng)的過程中暴露了無人駕駛汽車的一項(xiàng)主要弱點(diǎn)。谷歌汽車遭遇的首起造成人身傷害的交通事故并非由自身系統(tǒng)故障導(dǎo)致，而恰恰是因?yàn)槿藶殄e(cuò)誤。

當(dāng)無人駕駛汽車正確聽從指示等待交通信號(hào)燈時(shí)，一位心不在焉的司機(jī)開車撞了上去。即便配備了一系列精密復(fù)雜的傳感器，無人駕駛汽車仍無力避免這場(chǎng)意外。所幸，事故僅導(dǎo)致幾名乘客頸部扭傷，但人們因此得到了教訓(xùn)：無人駕駛汽車在與人類共享道路時(shí)易出事故。

盡管無人駕駛汽車系統(tǒng)精密，但目前還沒有針對(duì)其他道路使用者（人類）的備用方案。人類駕駛員們能相互溝通和遷就，但也會(huì)在駕駛過程中出無數(shù)小差錯(cuò)，而無人駕駛汽車目前完全無法應(yīng)付這些狀況。

應(yīng)對(duì)行人

從人類駕駛員與行人間的互動(dòng)來看，汽車制造商面臨著諸多道德倫理難題，但它們也頗具啟發(fā)意義。

無人駕駛汽車需了解行人的行為模式，同時(shí)模仿行人預(yù)想中人類駕駛員的反應(yīng)?！澳阄医匀祟?，對(duì)于人會(huì)作何反應(yīng)，我們都心中有數(shù)。”計(jì)算機(jī)倫理學(xué)評(píng)論員本·拜福德說，“因此如果你走到一輛車跟前，假設(shè)司機(jī)知道你在那兒，他會(huì)作出特定的反應(yīng)?！?/p>

“如果我闖到一輛正以每小時(shí)60英里車速行駛的谷歌無人駕駛汽車前方，由于無法按常理預(yù)測(cè)該車的行為，我實(shí)際上是使自己陷入了不利的境地?！薄?/p>

（譯者單位：北京航空航天大學(xué)）