Science And Technology Enterprises Show Their Muscles Frequently: Competition For Driverless Routes

In the last month of 2020, there is still a lot of news about unmanned driving this year.

"Finally we can not use the safety officer. It's a big responsibility." At 8 p.m. on December 7, Li Yanhong, chairman and CEO of Baidu company, updated his 100 brand dynamic. On the afternoon of that day, baidu received the first batch of unmanned road test notices in Beijing, which means that its relevant unmanned vehicles can cancel the safety officer during the test in the future.

Among them, the solutions of Apollo and Baidu smart cloud will attract more attention. The cancellation of lidar and the replacement of 10 line camera configuration also marks the official appearance of Baidu's pure vision application solution, which is lighter in sensor and calculation, stronger in perception, and, more importantly, lower in cost.

In addition to Baidu, other domestic driverless technology enterprises also spread news frequently. Although the Shenzhen traffic control bureau said that autox did not obtain its test license plate, the robotaxi motorcade, which was announced by the latter to be completely unmanned and without remote control, still swiped the screen on December 3. Wenyuan Zhixing recently released relevant data of its robotaxi, saying that it has completed more than 140000 trips in a year, and has served more than 60000 users.

"After a few years of technical research and development and preliminary measurement, I think this year and next year will be the stage of extensive road test for automatic driving," Pan Helin, executive director and professor of Digital Economy Research Institute of Central South University of Finance and law, told the 21st century economic reporter, "once the actual measurement stage is entered, the whole pattern of automatic driving industry will show great changes."

On the eve of the golden age

On December 3, at the 2020 World Smart car conference held in Guangzhou, Gao Jian, deputy director of the International Cooperation Department of the national development and Reform Commission, said that China's intelligent automobile industry is ushering in a golden period of development, and China will become the world's largest smart car market.

Gao Jian said that in terms of Internet of vehicles, it is estimated that the number of connected vehicles in the world will be close to 74 million in 2025, including 28 million in China. As for automatic driving, as of June this year, about 282 driving test licenses have been issued in 17 cities nationwide.

The attitude of the industry is also quite optimistic. In September this year, Li Yanhong even said directly that the unmanned driving technology will be commercialized on a large scale within five years. Cheng Wei, founder and CEO of didi travel, said that by 2025, more than 1 million shared cars equipped with autonomous driving will be popularized on didi platform.

J.P. Morgan Research and analysis said that from 2019 to 2025, China's L1 to L5 class auxiliary driving and automatic driving market size will achieve an average annual growth rate of 33%, and reach about $7.1 billion in 2025.

It is not only the improvement of digital technology, but also the road test with safety guards for a long time. The industry has finally begun to think about how to transform into a real "unmanned vehicle".

At the baidu World Conference in September this year, baidu released Apollo's "5g cloud driving" function: when an unmanned vehicle encounters a difficult roadblock, it can turn to "5g cloud driving" for help, and the cloud driving driver in the background control cabin can take over immediately and remotely control the vehicle to return to normal driving. Autox recently released a fully unmanned, non remote control robotaxi team, also marks the direction of this effort.

Gu DASONG, executive director of the traffic law and development research center of Southeast University, pointed out to the reporter of 21st century economic report that the depth of unmanned driving test and the number of test cities may reach a new level next year.

The so-called test depth includes not only the elimination of safety officers in the industry, but also the increase of test roads and test areas. "Next year, many cities will see the trend of automatic driving, and the test may be further launched. This trend is obvious."

Different technical routes

Vehicle road coordination is regarded as a promising route, and the domestic layout is also speeding up. "Vehicle road collaboration is very strong." An industry observer told reporters, "the single car sensor may be misidentified and misjudged due to various factors, but the accuracy of vehicle road coordination in grasping road conditions is much higher."

However, considering that the current vehicle road coordination is still in the planning and construction state, the industry is at a relatively early stage.

The longer developing cycle intelligent route emphasizes the ability of the vehicle itself. At present, the industry is divided into two types: the integrated perception mode and the visual recognition mode represented by Tesla. Companies including Google, Baidu and other enterprises adopt the integrated sensing mode. In addition to cameras, millimeter wave radar and ultrasonic radar, lidar is also added to the sensor. The pure vision mode mainly relies on camera perception, which completely simulates human vision.

From the perspective of perception principle, the way of lidar spot reflection can accurately judge the existence of objects, thus forming the basis for subsequent rapid driving decisions. The principle of the camera is to transform the two-dimensional image to the three-dimensional image, and then judge the volume of the object before making the driving decision. "This method requires high computational power." There are economic reports in the 21st century.

The person also pointed out that the observation distance of the visual recognition scheme is about 200 meters, while the lidar can realize 360 degree observation within 500 meters and achieve centimeter accuracy within 200 meters. Therefore, in terms of reliability, in order to achieve 99.999999% of vehicle safety, obviously the fusion perception mode is better.

However, compared with the visual recognition scheme, the fused perceptual mode is obviously more expensive. According to the source, the most expensive 64 line and 128 line lidar on the market may cost more than 800000 yuan per unit. The product price of domestic enterprise Hesai technology is at 10000 yuan level.

In contrast, the 360 degree look around camera in Tesla scheme is only tens of yuan to hundreds of yuan, the price range of millimeter wave radar is in the level of ten thousand or ten thousand, and the price of ultrasonic radar is as low as less than 10 yuan. In order to achieve vehicle safety, the total package of Tesla solution may be less than 10000 yuan, while the integrated sensing scheme needs several hundred thousand yuan.

Perhaps this is one of the reasons why baidu has further launched Apollo Lite lightweight solution, which means Baidu has also embarked on the path of pure visual recognition. According to reports, the scheme uses 10 cameras and has more than 30 deep learning networks, but the overall computing power is less than 30tops. At the same time, in the face of complex urban roads, the solution supports L4 level automatic driving.

"The launch of this scheme is actually to meet the needs of the main engine plant." In fact, Baidu's L2 + driver can't be close to the next level of driving in the 21st century. In fact, Baidu's L2 + driver can't fully use the technology of driving alone in the next century.

"Lidar has always been the main theme of unmanned driving because of its rich spatial information. However, it lacks the long-range resolution of RGB images and other defects, which is also a bottleneck restricting the development of the industry," commented pan Helin. "The transformation made by baidu this time is an innovative attempt, which not only overcomes the disadvantages of lidar, but also can significantly reduce the cost, from the current point of view This kind of look around scheme has a better application prospect. "

However, this does not mean that the fusion perception mode will be abandoned. "If visual recognition wants to fully and clearly identify the road conditions, it needs to improve the computational power, but the industry does not think that it is optimistic to achieve the safety of vehicle specifications." The aforementioned observers told the 21st century economic report that "if the cost of domestic lidar will be greatly reduced and even equivalent to the current visual recognition scheme in the future, the perception fusion mode will be further popularized."

?