Auto-driving cars face regional differences and unstructured driving becomes mainstream

Which areas most need the technology of self-driving cars? Recently, an article published in Wired pointed out that in the Middle East or Africa where accidents occur frequently, auto-driving car technology is most needed to reduce casualties. However, because of these really complicated traffic conditions, these areas are determined to be the least likely to obtain this technology in the short term.

Self-driving cars are no longer limited to controlled test roads, and are no longer limited to quiet suburban streets - they are responding to real traffic conditions in American cities such as New York, San Francisco, and Pittsburgh. They hone their skills in humans in Europe, South Korea, Singapore and Japan. They are preparing for the future. On that day, they can use perfect robots to keep our chaotic streets organized.

Learning how to drive in such an unruly place in Boston is very challenging. However, compared with the aggressive driving environment and complicated urban streets of developing countries, it is pale. Moreover, facing drivers who have no respect for lanes, traffic lights, warning signs and speed limits, it is also a real headache.

On a broad road without a driveway, the huge, unguided crossroads around the world, the interaction between people determines the flow of traffic, and each driver makes adjustments on the spot. They do not manage the rules manual at all. what.

These informal systems have played a role to a large extent, but they are costly. According to the data from the World Health Organization in 2013 (the latest available data), 50 of the 50 highway accidents have the highest risk, and 44 are in Africa or the Middle East. In 2013, the number of people killed in traffic accidents in these countries was nearly 250,000, accounting for one-fifth of the world total.

However, while these circumstances will make these areas most likely to benefit from self-driving cars, it also makes them least likely to acquire this technology in the short term.

"If we test in third-world countries, many of the things we do in autopilot may not work," said Ram Vasudevan, co-director of the Ford Autonomous Vehicle Center at the University of Michigan.

Unstructured driving

The good operation of automatic driving requires understanding the intentions and trajectories of everyone on the road, each object: vehicles, cyclists, pedestrians, construction workers, children, pets, and darts coming out of the Nerf gun. In the driving environment, people actually follow a set of rules that restrict the various behavioral intents that are predicted by autonomous vehicles.

The fewer formal rules, the more predictive power you need. In the vicinity of humans in remote areas, cars cannot rely on the so-called common principles to determine behavior. For example, if all other drivers on the road respect the markings on the lane, basic driving assistance functions allow the car to keep running smoothly in the lane. Otherwise they are useless or even dangerous.

Compared with suburban and even U.S. cities, the driving environment of many Middle Eastern and African countries has a structure similar to jellyfish. In Lebanon, it is common to see retrograde cars, red lights, tortuous roads on wide roads, no sign of lane markings, and other mischiefs.

"There is no rule here." "Everything is possible," said Daniel Asmar, a computer vision expert and professor of engineering at the American University of Beirut. "Humans can handle this problem well, even if they are frustrated and attack each other." For computers, this confusion will be a huge challenge.

Vasudevan said that even in a relatively orderly environment, a confusing situation, such as the parallel of highways, can make auto-driving cars hesitate for long enough to hinder traffic and even cause accidents. This may be because the programs in the car are not willing to be merged into the main road in front of the speeding car, or it may be because the car needs more time to understand the surrounding scenes and the intentions of other drivers. Compared to running on roads with parking signs, traffic signals, and giving rules, self-driving cars run on roads where rules or rules are often overlooked, and their response time needs to be more agile to survive.

What's more, self-driving cars require the help of map data that is not present in most parts of the world. Autopilot requires highly detailed street maps, from the height of street restrictions to temporary construction sites, to the precise location of street signs and traffic lights in 3D space. Cities with self-driving fleets are already developing these maps, and they are also using data captured by autonomous vehicles as they travel, constantly updating maps.

In places like Lebanon, two-dimensional maps of Google and Apple all contain basic mistakes, and losing data is a huge disadvantage. Even with detailed maps, they also require extensive maintenance. Asmar said: "In a structured environment, you don't need to do this often, because things are basically the same." In an unstructured environment, things are changing at any time, you can imagine that you How many times do you need to constantly build this platform? This is a very arduous task. ”

Some of the rich countries in the Middle East have begun to develop in the direction of autonomous driving. Israeli companies are an important driver behind the development of autonomous driving software. Last month, Israel opened the first test track for driverless cars. In Dubai, a 10-seater drone has begun to drive slowly in the Riverside Business District last year. City officials plan that by 2030, without a driver, one quarter of the area will be driven without a driver. The Dubai Police is also planning to launch a micro-driverless patrol car before the end of this year.

But it seems that only India and China will both cause chaos and there are also local companies that develop autonomous vehicles. As expected, their efforts face additional obstacles. According to Bloomberg, Tata India has set up a test track outside Bangalore to simulate local roads, including fearless pedestrians and stray cattle. The company still has a long way to go: Tata's senior vice president tells Bloomberg that its computer vision system currently does not recognize 15% of vehicles on Indian roads because of their diversity in shape and size. (When Uber's former chief executive, Travis Kalanick, visited India last year, he joked that the country will be "the last" unmanned car on the planet. "Do you see people driving cars?")

At the same time, China’s Baidu is also making its way into the field of automated driving and is cooperating with more than 50 international companies to develop its software. In a recent video presentation, Baidu’s chief executive, Li Yanhong, sat in a self-driving car and walked through Beijing’s traffic, carrying out several unsafe exercises along the way. Since auto-pilot cars are not yet legal in China, Chinese police said they would investigate whether Robin Li violated the law. (India is also taking a similar ban on grounds of fear of unemployment.) Despite regulatory hurdles, Baidu’s president, Zhang Yaqin, told Bloomberg that he believes Baidu’s self-driving car will be on its way to the next year.

China's leading taxi company Didi took a more cautious approach. Despite earlier this year, the company opened an office in California to develop autonomous driving technology, but the company’s president, Liu Qing, recently accepted Charlie? Rose said in an interview that, suddenly, "subversive" autopilot technology will be dangerous. "I think people should pay more attention to its security than how long it can come out," Liu Qing said.

A spokesman for Didi said that in China, self-driving cars not only need to learn how to deal with non-compliance cars, electric cars, and pedestrians, but they also need to be able to understand the regional differences between signs and traffic signals. These signals are not in China. Standardized as in the United States or Europe. There, the size of the drop gave it an advantage. The company said that its human drivers provide 25 million rides per day and generate more than 70 TB of data each day in order to develop its own autonomous driving capabilities.

regional difference

Currently, many companies are testing their self-driving cars by throwing unexpected scenes on controlled test tracks. At the Castle secret base in Waymo, human assistants cut off the roads of self-driving minivans at high speed, returning them from the blind lane to their roads, and throwing basketball at them, all to test and improve the car's response.

However, artificial intelligence trained under a set of assumptions may fail in different situations. The study found that face recognition algorithms based on white subjects performed poorly on African-American faces. Similarly, algorithms based on East Asian subjects also performed poorly on white faces. The same situation may occur in a self-driving car. The worst training situations are nothing more than a fly-by-time basketball or a self-driving car with similar conditions. When it is merged into the main road on a high-speed highway, it is very likely that it will be speeded up by the next two. The car gives "scared".

Despite the huge geographical differences in people’s driving styles, manufacturers may not need to create a Ghanaian version, an Iranian version, and an Indian version of the driving software. "In the context of every culture there will be the same mathematical principles and the same software," said Matthew Johnson-Roberson, a professor of engineering at the University of Michigan and another co-director of the Ford Center.

Most importantly, a well-trained, self-driving car responds to all vehicles. A spokesman for Uber, which is testing auto-driving cars in the United States and Canada, said that in order to improve the adaptability of the software, the company’s cars have traveled in multiple cities, under different conditions, and for different periods of time. One million miles.

Even if the autopilot software understands uncontrolled drivers and can predict that they may violate the law, auto-driving cars may also be limited. A spokesman for Uber stated that Uber’s car will always comply with local traffic regulations. Stephan Hoenle, Senior Vice President of Bosch Automation Drive, agrees. Hoenle said: "You can drive or defend more aggressively without violating the rules." The driving style of self-driving cars may vary according to needs and expectations, but breaking the law is not an option - for manufacturers, This is a huge responsibility.

The problem is that in some places driving according to legal provisions may be more dangerous than imitating illegal human drivers. If an impassive commuter turns a two-lane highway into a four-lane freeway at the peak of commuting, the self-driving car cannot make adjustments, which will soon lead to a series of crashes.

Who is the first?

For those technicians who train their computers every day to make them more advanced than humans, the details of the ultimate direction of self-driving cars do not seem so urgent. "It doesn't even take into account this aspect, right?" said Johnson-Roberson of the University of Michigan. "From an engineering point of view, I do not know who is studying this issue because some basic principles have not yet been realized."

Delaying consideration of these issues may put the region that needs the most autopilot technology to the last. Hoenle claims that the final of the driverless car will appear around the world, but he admits that this will not happen at the same time. He said that compared with the United States and Europe, "normally, the growth curve of other continents is slower."

Carlo Ratti, director of the SenseableCity Lab at MIT, predicts that developing countries will eventually catch up. In an e-mail he wrote, "Every technology needs to start somewhere - and often it starts with the most cutting-edge." "Initially, new technology can increase the existing social gap. However, technology Subsequent transmissions may bring about an interesting "crossing" effect and help close the gap."

In the case of mobile phones, only wealthy Westerners were able to use them. Nowadays, in Africa, startup companies are making efforts to emerge a large number of mobile banking and health care services. Ratti said: "There is no reason to believe that self-driving cars will take different paths."

For self-driving cars, the gap between the introduction and the "jump" phase may be much longer. Self-driving cars must adapt to the surrounding environment and need to have specific data on each street. If the design is poor, they may be “killed”.

Those developers who question regional differences and leave the issue to the “rising curve” will be locked out of a huge market. For now, as autonomous vehicles drive on friendly roads in places such as North America, Europe, and Singapore, it may leave developing countries most in need of such technologies behind.

COB Fine Pixel Pitch LED Display

Cob Led Display,Led Wall Pixel,Led Screen Pixel Pitch,Small Pitch Led Display

Shenzhen Apexls Optoelectronic Co., Ltd , http://www.apexls-display.com