On April 28, Tesla founder Elon Musk visited China. In the evening, the China Association of Automobile Manufacturers and the National Computer Network Emergency Technology Processing and Coordination Center issued the "Notice on the Testing of 4 Safety Requirements for Automotive Data Processing (the First Batch)", and 76 models from 6 companies, including BYD, Ideal, Lotus, Hezhong New Energy, Tesla, and Weilai, met the 4 compliance requirements of automotive data security (anonymization of face information outside the car, no collection of cockpit data by default, in-vehicle processing of cockpit data, and significant notification of personal information). Tesla is the only foreign-owned company that meets the compliance requirements.
As soon as the news came out, people in the industry seemed to hear the footsteps of Tesla's FSD entering China. According to people familiar with the matter, Tesla's FSD will be officially launched in the Chinese market this summer.
FSD enters China, are friends panicking?
He Xiaopeng, chairman of Xpeng Motors, was very generous this time, he welcomed FSD, and believed that whether it is FSD, or Xpeng's NGP, or NOA of friends, they are all representatives of new quality productivity and should support the blooming of a hundred flowers.
In fact, domestic car companies have begun to learn Tesla's purely visual route earlier.
If the pure visual route was only "the insistence of the few" and the lidar was the "choice of the majority", now the situation is quietly changing. A few days ago, Huawei and GAC Group successively released pure visual intelligent driving systems, and Robin Li, founder, chairman and CEO of Baidu, personally went to the car to demonstrate the pure visual intelligent driving function of Jiyue Automobile, and the battle situation of the autonomous driving perception technology route became confusing again.
01
New members continue to pour in, and pure vision is favored
The dispute between the pure vision technology route and the multi-sensor fusion technology route of "camera + radar" has been around for a long time. Previously, most domestic and foreign car companies such as Weilai, Xiaopeng, Great Wall, BMW, and Volvo have chosen the LiDAR route. Li Bin, chairman and CEO of NIO, has emphasized that lidar is indispensable for "responsible" high-level autonomous assisted driving. However, with the joining of a number of car companies and autonomous driving suppliers such as Huawei, GAC, and MAXIEYE Intelligent Driving Technology, the pure vision technology route has once again attracted attention from the industry.
A few days ago, Huawei officially released the HUAWEI ADS Basic Edition, a visual intelligent driving solution, which can now realize intelligent assisted driving and intelligent parking functions on high-speed urban express sections across the country. Described as "the best spot" by Yu Chengdong, Executive Director of Huawei and Chairman of Intelligent Vehicle Solution BU, Huawei's visual intelligent driving solution is equipped with three millimeter-wave radars, 10 visual perception HD camera groups, and 12 ultrasonic radars. On the premise of not relying on lidar and high-precision maps, it can realize NCA intelligent driving pilot assistance for national highways and urban expressways, including on/off ramps, autonomous obstacle avoidance, detour road construction, cone barrel diversion and other scenarios.
Also got rid of lidar and high-precision maps, as well as GAC Group. The day after Huawei released its visual intelligent driving solution, GAC's pictureless pure visual intelligent driving solution GARCIA also followed. The intelligent driving system does not rely on high-precision maps and lidars, but combines visual sensors and industry-leading AI algorithms to perceive the world like a human, and can achieve intelligent driving with a wider coverage with a simpler hardware system. According to reports, the perception range of GAC GAR-CIA reaches more than 150 meters in front of the car, 100 meters behind the car, and 50 meters around the car.
In addition to Huawei and GAC, Jiyue also released a pure visual high-end intelligent driving technology solution in October last year, which is known as the first pure visual intelligent driving solution in China that opens up the full scenario of "city + high-speed" "BEV+Transformer + OCC occupied network". It is reported that Jiyue has adopted a plan to promote the research and development and implementation of pure vision technology in stages. In the 1.0 stage, the intelligent driving scheme of "pure vision + lidar" integration is adopted, and the two sets of dual independent systems are backed up and complemented by each other; In the 2.0 phase, you can eliminate the use of lidar and gradually reduce your dependence on high-definition maps. It is understood that ROBO Drive Max has realized the PPA (point-to-point pilot assist) function in Shanghai, Shenzhen, Hangzhou, Shenzhen and Guangzhou.
In November last year, MAXI-EYE, a provider of autonomous driving technology products, released the BEV platform architecture - Qingyun Hyperspace, and announced the development of BEV perception standard parts for the entire industry. BENEFITING FROM QINGYUN'S BEV ARCHITECTURE, MAXIEYE HAS ALSO RELEASED A HIGH-SPEED AND URBAN NOA MASS PRODUCTION SOLUTION THAT DOES NOT RELY ON HIGH-PRECISION MAPS AND LIDARS, OPENING UP THE APPLICATION OF INTELLIGENT MEMORY DRIVING AND PARKING FUNCTIONS FOR THE FIRST DRIVE, RE-RECORDING AND TRIPLE ACTIVATION. According to the relevant person in charge of MAXIEYE Intelligent Driving Technology, MAXIEYE has added 2~3 mass production projects of leading OEMs this year, and the goal is to add at least 5 mainstream car companies this year. "Most of the new customers this year are focused on the NOA program. We expect the NOA market to usher in a round of explosive growth this year and next. He said.
In addition, in March this year, Shen Shaojie, the head of DJI Vehicle, released the urban high-end intelligent driving solutions "Chengxing Platform Upgrade" and "Chengxing Platform High-configuration Version". Among them, the upgraded version of the Chengxing platform is 7V+100TOPS, which only relies on 7 cameras and 1 Qualcomm chip 100TOPS computing power chip to achieve the function of intelligent driving in urban areas without pictures and pure vision. It is reported that SAIC Volkswagen and FAW-Volkswagen will adopt the inertial navigation binocular pure visual intelligent driving solution in the new Tiguan L Pro version and the new generation of Magotan in the new Tiguan L Pro version.
02
Technology + cost-effective to help pure vision "soar"
Dai Yifan, assistant dean of Suzhou Automobile Research Institute of Tsinghua University, told reporters that there are two main reasons why the pure vision technology route is suddenly highly respected.
First, the advancement of artificial intelligence technology has made the space available for the purely visual route larger. Previously, most of the industry believed that due to technical limitations, purely visual routes that rely on cameras are more likely to misjudge driving routes and distances in the face of extremely complex road conditions, light or environments, and cannot accurately describe the environment. However, Dai Yifan said that with the blessing of advanced technologies such as BEV+Transformer and occupancy network technology (OCC), the accuracy of pure vision technology has been greatly improved. "Under the application of these technologies, non-specific targets can be identified through pure visual methods, so that everyone can see the possibility of replacing lidar with vision." He said.
In addition to the significant improvement of perception accuracy, the relevant person in charge of MAXIEYE intelligent driving technology also said that autonomous driving, as a cross-field of AI and transportation, also has the characteristics of continuous learning and iteration of AI.
Second, the limitations of lidar itself in terms of cost and product maturity have also promoted the development of pure vision technology to a certain extent. "There have always been some doubts about the reliability and working life of the lidar itself. The life of lidar is certainly incomparable with that of cameras, and as an auto part, long life is a condition that must be met. Dai Yifan said.
Third, the cost disadvantage of lidar is more prominent. "Although the cost of lidar is declining, most of it is still more than 1,000 yuan, and the cost is still high for a car that costs only 100,000 yuan." Dai Yifan said. According to the research report of Huatai Securities, the cost of a single lidar has dropped to about 4,000 yuan. However, in the automotive market, where the price war is intensifying, the stacking of radars leads to high hardware costs, which is not in line with the current price reduction strategy of car companies. Yu Chengdong lamented that it is difficult for models below 300,000 yuan to use Huawei's lidar intelligent driving solution.
However, in the composition of the domestic passenger car market, the market price of vehicles is 80,000 ~ 200,000 yuan, accounting for more than 70%, and this huge market needs more cost-effective intelligent driving solutions. It is understood that the cost of DJI's 7V+100TOPS inertial navigation binocular pure vision intelligent driving solution is about 7,000 yuan, and the hardware cost of its 7V+32TOPS solution has reached the level of 5,000 yuan. MAXIEYE Intelligent Driving Technology has launched a product matrix of intelligent driving solutions for models at different price points, among which MAXIPILOT 2.0 Lite can support models below 200,000 yuan to achieve cost-effective light map high-speed NOA. A few days ago, Tesla reduced the subscription price of its FSD software from $199 to $99 per month, making the price of high-end intelligent driving more affordable.
In addition, Wang Yu, vice chairman and secretary general and researcher of the China Association of Productivity Promotion Centers, also pointed out that because Tesla has been deeply involved in the pure vision technology route for many years, it has also proved significant results to the industry in practical application.
03
Pure vision wins? It's too early!
"In terms of technology, AI and data-driven will push the vision technology route to new heights; In terms of cost, pure vision has significant cost advantages, and at the same time, it also has advantages in standardization and platformization. The relevant person in charge of MAXIEYE intelligent driving technology believes that the pure vision technology route is expected to become the mainstream route of high-end intelligent driving in the future.
In this regard, Dai Yifan believes that whether it is Jiyue, GAC and other car companies that have lost the two "burdens" of high-precision maps and lidars, or Xiaopeng, Weilai, Ideal, Zhiji and other brands that have started to "reduce the burden" from the loss of high-precision maps, they are closely related to the development of pure visual technology routes. Therefore, in the next few years, pure vision technology will inevitably be embraced by more car companies, especially in the relatively affordable mainstream passenger car market.
However, this does not mean that the battle for autonomous driving perception routes can be over. In Dai Yifan's view, the cost of lidar is expected to further decline, which will hinder the development of pure vision technology. "If lidar becomes cheaper, using it to occupy the grid will definitely be more accurate, more accurate, and more reliable than a camera. Therefore, in the future, it is necessary to consider which of the two technical routes has more advantages in terms of cost performance. He said. It is understood that Suteng Juchuang recently released a new lidar product MX. Qiu Chunchao, CEO of Suteng Juchuang, said that MX will start mass production of the first project based on about 200 US dollars, and then the price will strive to reach the level of 1,000 yuan as soon as possible. In recent years, the average unit price of Suteng Juchuang's ADAS application products has dropped from about 10,000 yuan in 2021 to about 3,200 yuan in 2023.
In addition, even if the technology of the purely visual route has made significant progress, the gap between domestic companies and the "originator" Tesla is not something that can be caught up in the short term. Whether it is the accumulation of data volume, the improvement of computing power, or hardware development and software research and development, Tesla has achieved many years of accumulation. In March this year, Tesla's FSD V12 was updated, and the end-to-end algorithms in the entire system are almost all built with neural networks, which are very flexible, adaptable, and can drive in a highly anthropomorphic state. "The country is still in a state of following, we still have to face up to our own gaps and try to keep up with the rhythm." Dai Yifan said.
"Whether it is lidar or pure vision, it is now applied to vehicles mostly based on L2 assisted driving, and the driver has not yet left the steering wheel at this stage, and the two routes can run through at the level of assisted driving. However, when it comes to the L4 or even L5 intelligent driving stage, at present, it is definitely not enough to rely on cameras or radars, but to use 'radar + cameras' in parallel. Wang Yu said.
Text: Zhang Yiwen Editor: Huang Bei Layout: Wang Kun
Add a chicken leg to the editor!