好英语网自动驾驶汽车如何开上印度拥挤的街道
“In 60 seconds you have to consider 70 options,” says my rickshaw driver Raju, leaning over his shoulder as we weave through traffic. We’re navigating the infamous congested streets of Bangalore, and he’s explaining the rules of the road.
"你得在60秒内考虑70种选择。"我的人力车司机拉朱（Raju）说，他边说边拉着我在路上穿梭。我们正在班加罗尔（Bangalore）臭名昭著的拥挤街道上前进，一路上听他给我讲解交通规则。
Having lived in India for two-and-a-half years, I get what he means. Not an inch of the road is wasted – if there’s a gap, a scooter will fill it. Vehicles travel bumper to bumper. Overtaking is attempted as frequently as possible. Indicators and wing mirrors are optional extras. Most drivers seem to rely on the incessant honking of nearby vehicles – almost a form of echolocation.
已经在印度住了两年半的我，完全能够理解他的意思。这里没有一点地方会被浪费——但凡有点空，就会有踏板车插进来。汽车一辆挨着一辆，几乎没有一点空隙。超车早就司空见惯。转弯指示灯和后视镜都是可有可无的东西。多数司机都凭借附近车辆响个不停的喇叭声判断路况——感觉就像回声定位一样。
But there is method to the madness. Drivers deftly navigate around manoeuvres that would lead to accidents in the UK, and offenders rarely elicit more than a mutter. They’ve adapted to predictable unpredictability.
但这种疯狂的道路并非无迹可寻。如果换成是英国，在车流中如此穿梭肯定会引发事故，但在这里，受到冒犯的人最多只会抱怨两句。这些杂乱无章完全在意料之中，他们已然习以为常。
印度拥挤的街道就像一个包含了不同路障（甚至包括大象）的独特万花筒，可以为无人驾驶人工智能技术提供更好的数据
"任何事情都有可能发生，只有这样才能进行预测。我们一直都预计前面的汽车会左右转向或者突然刹车。"拉朱说，"有些事情总会发生。"
This driving philosophy has complex cultural and historical roots, but it's also a product of rapid growth in both urban populations and vehicle ownership – which government figures show nearly quadrupled between 2000 and 2013. This growth is outstripping the ability to build new infrastructure, leaving citizens to adapt.
这种驾驶理念有着复杂的文化和历史渊源，但城市人口和汽车保有量的快速增长也是造成这种现状的原因之一——官方数据显示，2000至2013年间，印度汽车保有量增长了接近3倍。这种增长速度远超印度的基础设施建设速度，所以市民只能被迫适应。
India is far from alone – rapid urban overcrowding and car ownership put countries like China, Brazil and the Philippines all in a similar boat.
印度并非个例——城镇人口和汽车保有量的快速增长导致中国、巴西和菲律宾等国家都陷入了同样的窘境。
But for many tech companies and researchers, India’s chaotic roads make it the perfect testing ground. They might help us solve some of the big problems that come with living in crowded cities… ones that are only going to keep getting more crowded.
但对很多科技公司和研究人员来说，印度混乱的街道反而成了完美的测试场。这里或许可以帮助我们解决拥挤的城市里最为严重的一些问题……这些城市今后的拥挤程度只会有增无减。
***
Overcrowding in Earth’s cities is a complex, nuanced problem with no magic bullet solution – it requires a comprehensive game plan that addresses everything from infrastructure to energy to income inequality.
城市的拥挤不堪是个复杂而微妙的问题，没有什么灵丹妙药——需要制定一套全面的策略来解决包括基础设施、能源和收入不平等在内的各种问题。
But as far as those cities’ roads are concerned, technologists all over the world are pursuing one grand idea in particular – one idea that doesn’t even involve humans.
尽管这些城市的道路状况堪忧，但世界各地的技术人员都在追求一个宏伟的想法——这个想法甚至完全不需要人类参与。
Self-driving cars bring the promise to keep traffic flowing and to help us optimise our journeys. The argument is that smart robot cars could communicate with each other, better sense obstacles and generally create a more organised flow of traffic.
自动驾驶汽车号称可以保持车流顺畅，优化交通效率。支持者认为，这种智能机器人汽车可以相互沟通信息，还能更好地感知障碍物，因此整体而言能够营造更加井井有条的交通状况。
But much of this technology is being developed for the orderly streets of the West. Uber’s former CEO Travis Kalanick has said India would be the last place to get driverless cars after experiencing Delhi traffic.
但这项技术主要是针对西方老旧的街道开发的。Uber前首席执行官特拉维斯·卡兰尼克（Travis Kalanick）在体验了德里的交通后表示，印度将是无人驾驶汽车最后覆盖的地方。
That hasn’t deterred design firm Tata Elxsi, which is building a self-driving system that could be retrofitted to any car, and have started testing prototypes on a test track near their Bangalore headquarters.
但设计公司Tata Elxsi并未因此气馁，他们正在开发能改装任何汽车的无人驾驶系统，并且已经开始在其班加罗尔总部附近的一条试车跑道上测试原型车。
Road testing is far away though and they’re under no illusions about the challenges. “Driving behaviour is a lot more erratic here,” says Nitin Pai, head of marketing and strategy. “Rules are not rules, they're more guidelines.”
但路测依然遥遥无期，他们对未来的挑战也并未心存幻想。"这里的驾驶行为很不规则。"该公司营销和战略主管尼汀·帕伊（Nitin Pai）说，"规矩不成规矩，反倒更像是一种参考。"
And that’s exactly why thinking about Indian roads in particular could help us come up with the best ideas.
正因如此，多多思考印度的道路才能帮助我们想出最好的创意。
Self-driving cars rely heavily on machine learning – that’s when AI uses mountains of data to train itself to do things like recognise vehicles and predict their trajectories over time. So far, such cars have been tested across the globe in places like San Francisco and Pittsburgh in the US, and in smaller cities in countries like Japan and China.
无人驾驶汽车高度依赖机器学习技术——在这种情况下，人工智能需要使用海量数据进行自我训练，从而慢慢地学会识别车辆、预测轨迹。目前为止，这些汽车已经在世界各地展开测试，包括美国的旧金山和匹兹堡，以及日本和中国的一些小城市。
But most research has been done in the West, where driver behaviour could be argued to be more predictable and roads are reliable with clear signs.
但多数研究都来自西方，那里的驾驶行为更容易预测，清晰的道路标识也更加可靠。
"These are the foundations on which you lay the system of a driverless car,” says Pai.
"这是你开发无人驾驶汽车的基础。"帕伊说。
The group has been training their system on third party data collected by researchers at the Karlsruhe Institute of Technology in Germany. This was done on German roads using a car equipped with high resolution cameras, GPS and a Lidar sensor – effectively radar using laser light instead of sound for very sensitive distance measurements.
该团队一直在使用德国卡尔斯鲁厄理工学院（ Karlsruhe Institute）的研究人员收集的第三方数据来训练系统。这些数据都来自德国道路上行驶的汽车配备的高清摄像头、GPS和Lidar——它本质上是一个用激光代替声波的雷达，可以非常敏感地测量距离变化。
They’ve developed algorithms using this data, says head of strategic initiatives Rajesh Kumar, but machine learning is only as good as the data you give it. And if we want to solve road problems in overcrowded developing countries, we need data from those environments.
该公司战略项目主管拉杰什·库玛（Rajesh Kumar）表示，他们使用这项数据开发了一些算法，但机器学习的潜力却要受制于数据的好坏。如果我们想要在过度拥挤的发展中国家解决道路问题，那就需要在这些环境中收集数据。
Indian roads throw up plenty of obstacles that don’t appear on German roads, which means that an AI trained on German data won’t be able to recognise all the objects a self-driving car will encounter on India’s roads.
印度的公路上会出现很多在德国见不到的障碍物，因此，使用德国数据训练的人工智能无法识别一辆无人驾驶汽车可能在印度街道上遇到的所有物体。
For example, “any dataset you take from a European or American university won't feature a typical autorickshaw,” he says.
例如，"你从一所欧洲或美国大学获得的数据集不会包含嘟嘟车。"
“If there aren’t autos in my dataset my self-driving car will never recognise one because my AI system has not been trained to recognise autos.”
他说，"如果我的数据集里没有汽车，我的无人驾驶汽车就永远认不出汽车。原因在于我的人工智能系统没有接受过这种训练。"
It’s not just rickshaws; in India scooters and motorbikes are far more prevalent and cars also compete with a menagerie of unconventional road users.
不光是嘟嘟车；踏板车和摩托车在印度更加流行，汽车必须与很多不同寻常的道路参与者展开竞争。
“People suddenly come into the road, so do cattle, monkeys, elephants,” laughs Raju the rickshaw driver. “They are also part of our traffic so we are always looking out for them.”
"人们会突然之间闯到路上来，牲口、猴子和大象同样如此。"拉朱笑着说，"这都是我们交通的组成部分，所以我们总得留意他们。"
And all that could help self-driving car technology be better and safer in a way it couldn’t be without Indian road data.
有了印度的数据，无人驾驶汽车技术就能变得更好、更安全。
Capturing these differences is essential if self-driving cars are ever to work on Indian roads, says Kumar, but it’s also a data gold mine for researchers looking to perfect the technology for other developing countries.
库玛表示，如果无人驾驶汽车想要在印度的街道上行驶，就必须捕捉这些差异。不仅如此，对于那些想要针对其他发展中国家完善这项技术的研究人员而言，这里也是一个数据金矿。
So several times a week they drive a sensor-laden sedan around Bangalore’s Whitefield suburb to build up a local dataset of high-definition video, LIDAR data and high-precisions GPS readings.
所以，他们每周都会开着装载传感器的轿车到班加罗尔的怀特菲尔德（Whitefield）郊区转几圈，借此构建一个涵盖高清视频、Lidar数据和高精GPS信息的数据集。
Elxsi thinks their first customers will be in developed countries, but setting the bar high could eventually let them tap emerging markets with similar traffic conditions.
Tata Elxsi认为，他们的第一批客户将来自发达国家，但提高标准可以使之最终得以进入拥有相似路况的新兴市场。
"If we're able to solve Indian roads, we can do any roads,” says Kumar.
"如果我们能应付印度的道路，那任何道路都不在话下。"库玛说。
***
They aren’t the only ones trying to tame India’s byways and highways – which could then help cities run more smoothly worldwide. Lelitha Devi is an associate professor at the Indian Institute of Technology (IIT) Madras in Chennai, where she works on so-called intelligent transport systems (ITS).
他们并非唯一一家渴望驯服印度混乱路况的公司——这些信息还可以帮助世界各地的城市更加顺畅地运行。莱利塔·戴维（Lelitha Devi）是印度理工学院（IIT）的一位副教授，她在那里研究所谓的智能交通系统（ITS）。
This is the technology behind services people take for granted in the UK like journey planners, digital bus arrival time signs and smart control of traffic lights that keep traffic flowing.
这项技术为很多服务提供了支撑，包括英国人习以为常的行程规划工具、公交到达时间提示器和保障交通顺畅的信号灯智能控制器等。
Indian city administrators are increasingly looking to introduce similar technology, says Devi, but a lack of home-grown solutions means they often turn to Western providers.
戴维表示，印度城市管理者也越来越希望采用类似的技术，但由于缺乏国产解决方案，他们往往需要向西方供应商采购。
"Just because it works in the US or UK doesn't mean it will work in our cities,” she adds. "The characteristics are completely different, the complexity is much more. So we need indigenous solutions.”
"单纯因为这项技术适用于美国或英国，并不表明它也适用于我们的城市。"她补充道，"特点完全不同，复杂度也要高得多。所以我们需要土生土长的解决方案。"
Her lab has set up a network of video cameras and sensors along one of Chennai’s commuter corridors. Low-power wireless communication technology streams data to a makeshift traffic monitoring room at IIT to help them start building models of Indian traffic.
她的实验室已经在钦奈（Chennai）的一个上下班通道上建起了摄像头和传感器网络。低功耗无线通讯技术将数据发送到印度理工学院里临时设立的交通监控室，帮助他们开发印度的交通模型。
"In western traffic, there is one vehicle behind another and they all leave minimum spacing during congestion. Here, there will be vehicles in any space you have,” she says.
"在西方的交通环境下，车辆都会在拥堵时留下最小空隙。但在这里，只要留下空隙，就会有车插进来。"她说。
Some of her students are trying to model two-wheelers moving around vehicles using analogies from how fluids flow through air pockets in sand.
她的一些学生试图参照液体在沙子中流经气窝的方式来建模，借此模拟在汽车周围穿梭的两轮车。
Meanwhile, one of their models is already delivering results using GPS data from city buses to predict arrival times. Similar approaches are used in Europe, says Devi, but they tend to assume uniform speeds on subsections of the route.
与此同时，他们的一个模型已经可以借助城市公交车上的GPS数据预测到达时间。戴维表示，类似的方法已经在欧洲应用，但这些技术假定汽车在固定路径的各个路段中保持一致的速度。
“Here, you can't assume anything will be uniform or use any prediction method that does,” she says.
"在这里，你不可能假定任何东西保持一致，也不能使用任何持有这种假定的预测方法。"她说。
Even capturing data in the first place has required creative thinking. A popular approach to monitoring traffic is inductive loops. Wire is laid under each lane, and when a vehicle passes above, the wire’s magnetic field is disturbed, causing a signal to be transmitted to a counter.
就连收集数据也要融入一些创造性思维。感应线圈是监控交通的常见模式。每个车道下面都埋有线圈，当有汽车经过时，线圈的磁场就会被扰乱，从而发出信号。
The approach works when vehicles stick to their lane, but that’s rarely the situation in India. So Devi’s group built a system that accurately detects eight vehicle types even when passing simultaneously.
当汽车保持在同一个车道时，这种方法的确有效，但印度很少出现这种情况。所以，戴维的团队开发了一套能够精确探测8种车型的系统，就算它们同时经过也可以正常探测。
***
These projects are impressive, but they are in their infancy, and it will be hard to convince city planners to implement them.
这些项目都给人留下了深刻的印象，但也都处于发展初期，所以很难说服城市规划者部署这种技术。
Devi says officials often prefer established foreign solutions, regardless of whether they are appropriate for Indian conditions.
戴维表示，政府官员往往更偏爱已有的外国解决方案，而不考虑那些方案是否适合印度的环境。
But there is considerable impetus for this kind of technology. One of Prime Minister Narendra Modi’s flagship initiatives is the $15bn (£11.5bn) Smart Cities Mission aimed at using data to transform everything from transport to sanitation. The benefits could be considerable on the increasingly choked streets of India’s booming cities.
但这种技术的确有很大的发展动力。150亿美元（115亿英镑）的"百城大改造"（Smart Cities Mission）项目是印度总理纳伦德拉·莫迪（Narendra Modi）推动的重要项目之一，他希望借助数据改变交通和卫生等方方面面。对于印度新兴城市日益拥挤的街道来说，由此带来的好处十分可观。
Raju the rickshaw driver says 15 years ago, the average speed in Banglaore was 25mph (40km/h) , but today it’s down to around 6mph (10km/h).
人力车夫拉朱表示，15年前，班加罗尔的平均车速是25英里/小时（40公里/小时），现在大约只有6英里/小时（10公里/小时）。
"These new [autorickshaw] drivers starting now won’t last more than five years, because of the heavy traffic, the pollution. It will risk their health,” he says.
"由于交通拥堵和污染加重，现在才开始干的（嘟嘟车）驾驶员坚持不了5年。这会威胁他们的健康。"他说。
Devi thinks technology like their bus prediction system could help ease the burden by encouraging use of public transport. In passenger surveys, they found many would prefer to travel on busses, but simply can’t rely on them.
戴维认为，像他们的公交预测系统这样的技术可以鼓励人们使用公交车，从而减轻道路负担。他们对乘客进行调查后发现，很多人都喜欢乘坐巴士出行，但却无法依赖这种模式。
Her group is also working on route planning systems that could optimise road use. Their traffic predictions could one day feed into systems controlling traffic lights intelligently, based on traffic flow, rather than just using timers as done presently.
他的团队也在开发路径规划系统，可以优化道路使用效率。他们的交通预测技术有朝一日也可以融入交通信号灯控制系统，从而根据车流状况智能控制信号灯，而不再像现在这样单纯使用计时器。
But all the technology in the world could only help sort the wild roads of India – or anywhere else – so much. The best way to make the best use of a difficult driving landscape might be starting with the actual humans behind the wheel.
但无论是对印度还是其他地方而言，所有的技术所能发挥的作用也仅限于此。想要充分利用艰难的驾驶环境，最好的切入点或许还是方向盘后面的驾驶员。
Some transport engineers think trying to change the habits of India’s drivers is a better approach, but Devi feels that is beyond their control.
有的交通工程师认为，努力改变印度驾驶员的习惯或许更加可取，但戴维认为，这并非他们所能控制。
As with most things in India, it’s easier to go with the flow.
与印度的多数事情一样，人们很容易随大流。
“This is how the system is. The question is how do we improve it,” she says. “Anyway, it’s more efficient in some ways. We don’t waste any space on the road!"
"系统就是这样运作的。问题在于我们如何改进。"她说，"总之，它的确在某些方面效率更高。我们不会浪费任何道路空间！"
