China Naming Network - Eight-character query< - What is the biggest problem that restricts the development of autonomous driving?

What is the biggest problem that restricts the development of autonomous driving?

The Big Bang is coming: What is the biggest problem that restricts the development of autonomous driving?

The era of autonomous driving is coming.

20 18 is a big year for autonomous driving. After two years of continuous preheating of unmanned vehicle videos and various news, autonomous driving startups have mushroomed. Mutual benefit network giants and traditional car companies are unwilling to lag behind, and have announced the road test and landing schedule. Recently, a research report released by the Brookings Institution, an American research institution, pointed out that global manufacturers have invested 80 billion dollars in the field of autonomous driving. If you add those secretive autopilot research and development plans, the total investment should far exceed this figure.

As a result of multi-party competition, the time nodes of self-driving vehicles on the road are becoming clearer and closer. Since mid-June of 10, Waymo, owned by Google, has been operating L4-class fully automatic cars on public roads in Arizona, USA. The whole car is not monitored by the driver and is completely controlled by the automatic driving system.

Baidu, Google's rival in China, announced that it will mass-produce L4 driverless micro-circulation cars next year. This car can be operated in specific scenes, and can also cope with various actual sudden road conditions.

Many startups regard next year as an important node. Chi Jing Technology, founded by Wang Jin, said that the road test of driverless cars will be conducted in China in the first quarter of next year. In addition, it is reported that China may issue the first unmanned test license next year, which is very good news for various research institutions.

Autopilot is not a new concept, but its classification is not widely understood by the public. SAE International (International Association of Automotive Engineers) defines the participation of drivers, and divides the automatic driving from L0 to L5 into six grades. Among them, L5 is fully automatic driving in the whole scene, and L4 is an unmanned vehicle with a designed application range, that is, fully automatic driving in a specific scene. L2 and L3 are assisted driving and semi-automatic driving respectively.

Due to the balance between research cycle and commercialization prospect, most R&D teams still aim at L3 to L4, from semi-automatic driving with driver assistance to unmanned driving on fixed routes.

Two routes: Google and Tesla

Since the birth of the vehicle, people have had the idea of making it run automatically, but the real realization is due to the breakthrough of robot technology in the 1980s. After entering the 2 1 century, with the rapid development of computers, maps, sensors and automotive electronics, the research on driverless technology has ushered in an explosive period.

DARPA Grand Chanllenge, translated into Chinese as DARPA Driverless Technology Challenge, is one of the most influential events in this field, from which not only industry celebrities including sebastian thrun emerged, but also stimulated a series of related industry companies including Lidar company Verdun.

In the 2007 DARPA final, Google founder Larry Page came to the scene by commercial plane. He is eager to find a new direction of innovation for Google, and driverless is his first project. Two years later, Google began to develop unmanned vehicles. He introduced many researchers from the DARPA challenge champion team in 2005 and 2007, including Tron and Chris Urmson, who became the first and second leaders of the project. Google's move opened the prelude for Internet companies to enter the field of autonomous driving.

Looking back, the economic crisis in 2008 should be the key reason why Google's driverless research stood out among many car companies. At that time, many American automobile companies were forced to cut research funds. For example, General Motors, a major corporate sponsor of Carnegie Mellon University, filed for bankruptcy protection in 2009. Another important reason is the misjudgment of automobile manufacturers. These CEOs generally believe that this technology may develop in 2030-2040, and Google's first effort has won a good position for itself.

Google's driverless prototype was tested on Mountain View Road.

In the first five years of 20 10, Google almost became the spokesperson of unmanned vehicle black technology. However, with the stagnation of Google's own commercialization process, latecomers have begun to catch up. The most important event is the mass production of Tesla's self-driving vehicles.

Tesla is different from Google. Google chose to directly develop L4-level fully autonomous driving technology in one step, while Tesla took a step-by-step approach, starting with L2-level assisted driving. In June 5438+10 last year, Musk announced that all Tesla being produced would have completely unmanned functions, including Model S, Model X and Model 3, but this is only the hardware level. On the system level, the optional package of "enhanced automatic assisted driving" provided by Tesla to car owners still belongs to L2-L3 level. In the future, after the development of Tesla's L4 technology is completed, the subsequent optional packages will continue to be pushed to the owners to realize L4-level automatic driving.

2065438+In May 2006, Tesla's strategy was severely hit. Joshua Brown, the owner of Tesla in Ohio, USA, was too relaxed during driving. He took L2 as a fully automatic driving L4, fell asleep during driving, and finally hit a heavy truck, causing the car to be destroyed and killed. This accident caused people to question L2 technology, and the German government even asked Tesla to delete the misleading word "Autopilot" from the related publicity of this technology.

When media representatives visited Waymo last month, CEO John Crafts told the media that Google had completely given up assisted autonomous driving, because drivers would lose situational awareness during autonomous driving and it was difficult to take over the car when danger came.

In China: Baidu pre-empted, Tencent joined the battle.

When Google's unmanned vehicle was driving in California, Baidu, as a search giant in China, began to study deep learning and autonomous driving.

This R&D project began in July of 20 13, and it was not until one year later that Baidu confirmed the launch of the R&D plan of "Baidu Driverless Car".

On 20 15 12, Baidu officially announced the establishment of an autonomous driving division in Zhongguancun Software Park International Conference Center. It plans to realize the commercialization of autonomous vehicles within three years and mass production within five years. At the same time, BMW cooperated with Baidu, and the self-driving car developed on the basis of BMW 3 Series GT successfully carried out on-the-spot road test without driver intervention in Beijing Fifth Ring Road.

At the 20 16 Baidu World Congress, Wang Jin, then senior vice president of Baidu and head of the autonomous driving division, announced that Baidu had been awarded the 15 unmanned vehicle test card in the world by the California government. At this time, Baidu unmanned vehicle began to become the business card of China autonomous driving research.

With Wang Jin, Yu Kai, Peng Jun, Tong Xianqiao and other Baidu R&D personnel leaving their jobs to set up their own autonomous driving companies, the Baidu R&D team finally started a comprehensive commercialization effort on 20 17.

20 16 1 1, Baidu unmanned vehicle appeared in Wuzhen, world internet conference.

2065438+In March 2007, Baidu announced that it would integrate the Autopilot Division (L4), the Smart Car Division (L3) and the car networking business, and set up a special intelligent driving business group (IDG), with Qilu, president and chief operating officer of Baidu Group, as the general manager. In April, the "Apollo Project" was released, and Baidu announced to provide partners with an open, complete and secure software platform to help them quickly build their own complete autonomous driving system by combining vehicles and hardware systems.

This plan has attracted a large number of industrial chain partners for Baidu, and the commercialization of autonomous driving research has finally seen the dawn. In operation, Baidu chose to cooperate with the first car about the car. Next year, L3 intelligent driving team will operate in some cities. In the production of unmanned vehicles, the unmanned minibus "Apolon" jointly produced by Baidu and Jinlong Automobile will also be mass-produced in July next year. This car has neither a steering wheel nor a driver's seat. It is a real driverless car.

From the practical point of view, Baidu has taken L3 and L4 routes at the same time: on the one hand, L3' s assisted driving system can cooperate with most automobile OEMs to quickly realize the commercialization of smart cars; On the other hand, completely driverless cars are bound to be the ultimate direction of future car research and development, and Baidu, which has a lot of technology, has also broken the edge in this respect. Although different from Google's strategy of only taking one route, the strategy of coexistence of two routes also shows that Baidu has not achieved fully automatic driving through L3 technology upgrade.

It seems that the technical feature of China Company is to adhere to a two-pronged strategy. Tencent, another internet giant in China, also holds this attitude in the research and development of autonomous driving.

Tencent started three years later than Baidu. Last month, Su Kuifeng, director of Tencent Autopilot Lab, publicly acknowledged the existence of the Autopilot R&D project at Tencent Global Partner Conference and announced some progress. Su Kuifeng said that at present, Tencent pays more attention to L3 scheme in semi-closed expressway environment, and will also carry out algorithm research and data accumulation for L4/5 core technology.

The biggest problem facing autonomous driving

Autopilot faces too many technical problems. Sensing equipment, such as cameras, lidar, radar, etc. It needs to be more efficient, especially how to improve the resolution efficiency of sensors under bad weather conditions, which poses no small challenge to hardware and algorithms.

The cost of bicycles is also a big problem. In order to ensure safety, multi-system redundancy design is essential. For consumer-grade unmanned vehicles, the industry generally believes that multi-sensor fusion is the basic guarantee. Now the price of lidar is on the high side, and the price of vehicles is still too high for most consumers.

However, the technical and cost problems are obvious, which will be gradually solved with the evolution of the algorithm and the mass production of vehicles. The social cost involved in the popularization of autonomous driving is the biggest problem it faces.

Social cost should be considered from both physical and legal aspects. On the material level, infrastructure construction needs a lot of upgrading. Previously, the German government has made plans to realize the wireless communication function between vehicles and road infrastructure for the development of autonomous driving on every highway in Germany. In addition, the communication between vehicles can be realized by sensors and wireless devices installed on the roadside, which are the driving environment needed by self-driving cars.

In the era of unmanned driving, V2X capability, that is, the interaction capability between vehicles, vehicles and base stations, and vehicles and pedestrians, is also very important. This puts great demands on the construction of the Internet of Vehicles, the popularization of IOT and the 5G wireless network. These are also the comprehensive costs that society needs to pay.

V2X scene in the era of autonomous driving

In addition to wireless communication equipment, the smoothness of the road itself and the recognizability of the lane line are the keys to whether the self-driving car can drive safely. If these problems can be easily solved on the main roads in the central area of the city, it is not easy to realize the conditions of automatic driving or even unmanned driving on the secondary roads, branch roads and even suburban and rural roads.

In addition to these tangible costs, "intangible costs" invisibly restrict the development of autonomous driving. From a self-driving vehicle to an urban road, the existing traffic regulations are obviously far from meeting the needs.

On July 6 this year, in the live broadcast of Baidu AI Developers Conference, the self-driving vehicle developed by Baidu, which Li Yanhong was riding, violated the rules in "full view", the solid line changed lanes, and the turn signal did not turn on. After that, the Beijing traffic control department issued the first ticket for autonomous driving to Baidu.

For Baidu, it is meaningful to get this ticket. After all, the car is its own, the system is developed by itself, and the person in the driver's seat is also the general manager of Baidu Smart Car Division. But if it is a vehicle delivered to the user, who should this ticket be sent to? Is it the automobile main engine factory that delivers the vehicle, or the developer who develops the whole automatic driving system, or the user who owns the vehicle but has no driving behavior?

The Geneva Convention on Road Traffic (1949) requires drivers to "be able to control their own vehicles at any time", while the provisions on reckless driving usually require "consciously and purposefully operating vehicles". How should this rule be applied in the era of fully automatic driving?

Bryant Walker Smith, a professor of law at Stanford University, once wrote a document and put forward suggestions on how to adjust the law in the context of autonomous driving, including changing the word "driver" to include computers without eyes or ears in the conventional sense. The difficulty in amending the law is that the law can require human beings to regulate their own behavior, but now it can't require an artificial intelligence system to do anything, unless the lawmaker can clearly understand what the artificial intelligence system can and can't do, and judge who these responsibilities belong to under the complex technical mask. It is conceivable that with the advent of the era of autonomous driving, the existing traffic regulations will also usher in major changes.