news

Recently, foreign media reported that Tesla's FSD (Full-Self Driving) may encounter obstacles in entering China. Due to concerns about accidents caused by Tesla's software in the United States and data security issues, Chinese approval authorities have not yet approved Tesla's latest autonomous driving technology.

Regarding the above rumors and the time point for Tesla FSD to enter China, Tesla China responded to China News Weekly: "There is no relevant information."

In April this year, Tesla CEO Elon Musk made a "lightning visit to China" to promote the implementation of Tesla's FSD system in China. In the second quarter earnings conference call, Musk also stated that Tesla is expected to obtain FSD licenses in other markets, such as Europe and China, by the end of this year.

"Tesla FSD is a revolutionary technology and a great encouragement to the industry." At the 4th Shenyang Intelligent Connected Vehicle Conference held recently, Li Keqiang, an academician of the Chinese Academy of Engineering, affirmed Tesla FSD. However, Li Keqiang also said that Tesla FSD is actually still a high-level assisted driving, which is a "shadow mode" + "end-to-end large model". The complete data base required by the large model in the true sense, the integration of vehicle, road and cloud has a natural advantage. "The new generation of FSD and the integration of vehicle, road and cloud are not contradictory," said Li Keqiang.

In 2019, Tesla realized localized production in China, becoming a catfish that stimulated the development of domestic new energy vehicles. Now, Tesla FSD is about to enter China, and it may once again become a catfish that stimulates the development of domestic intelligent driving.

The wolf is coming? Opportunities and challenges coexist

As early as 2016, Musk had envisioned a bright future for Tesla owners: the car would drive automatically to take people when needed, and the car would go out to pick up jobs and make money when not needed.

Eight years have passed, and although the future that Musk envisioned has not yet truly arrived, the imagination space that intelligent driving brings to the industry is constantly growing and developing.

Image source: Tesla

"Intelligent driving, including complex systems integrating vehicles, roads and clouds, is about to enter the stage of large-scale application." According to Li Keqiang, in the first five months of this year, more than 50% of new cars in my country were equipped with intelligent driving assistance systems, and L3 level cars entered the quasi-commercialization stage, including the expansion of the demonstration scope of L4 level cars.

Zhu Xichan, a professor at the School of Automotive Engineering at Tongji University, said: "Tesla's end-to-end and artificial intelligence make autonomous driving more possible. Tesla has made end-to-end popular, and domestic car companies are all talking about end-to-end." However, in Zhu Xichan's view, domestic companies still have a big gap with Tesla in terms of data volume and computing power for AI training. "Xpeng Motors' AI computing power is relatively strong in China, but it is still dozens of times worse than Tesla."

Regarding Tesla FSD's upcoming entry into the Chinese market, Zhao Ji, secretary of the board of directors of BAIC Blue Valley, told China News Weekly: "What is coming will come, and this will be another test of the technological strength of domestic enterprises." In the first half of this year, Wuhan Carrot Run attracted a lot of attention, and Polar Fox also received a certain amount of attention. In the public notice released by the Ministry of Industry and Information Technology in June this year, Polar Fox Automobile was included in the first batch of pilot lists for access and road access to intelligent connected vehicles. "At present, Polar Fox is equipped with Hongmeng OS intelligent cockpit and Huawei driving assistance system in the field of intelligent cockpit and intelligent driving, and jointly copes with the intelligent transformation of the automotive industry with strong players." Zhao Ji said.

Du Qiang, president and CTO of Neusoft Reach, told China News Weekly: "The entry of Tesla FSD is a catfish effect. It is a challenge to our entire domestic autonomous driving, but also an incentive. In the long run, it is a very good opportunity to improve the overall autonomous driving capabilities."

"Faced with challenges, different car companies have different technical routes. Some, like Tesla, build pure vision solutions and move towards end-to-end solutions. The challenge of this route is how to build such a large-scale data center; some car companies insist on high-cost lidar." Du Qiang specifically mentioned the Chinese solution of vehicle-road-cloud integration. "Tesla is based on single-vehicle intelligence + cloud-based super-computing power. The vehicle-road-cloud integration solution makes more use of some road-side information and equipment to complement it. It can realize large-scale scene applications of autonomous driving in major cities, major trunk highways, and expressways with relatively good road coverage."

FSD is not in conflict with “vehicle-road-cloud integration”

To achieve autonomous driving, it is generally believed that there are two technical routes, one is FSD and the other is called "vehicle-road-cloud integration", but the two are not completely opposite.

Simply put, Tesla FSD achieves autonomous driving by installing eyes and brains in the car, allowing AI to learn to drive by seeing the road. The integration of vehicle, road and cloud can be understood as having signal transceivers on the vehicle side and on the roadside, which then transmit the signals to the cloud, and use the cloud to command vehicles and intersections for autonomous driving.

Li Keqiang said that FSD can enter assisted driving mode on a variety of roads, but due to limited scenarios, there are perception blind spots and beyond-visual-range perception limitations, and the urban MPI (takeover mileage) is still relatively short. It is actually a high-level assisted driving.

Li Keqiang believes that Tesla's FSD is a "shadow mode" + "end-to-end large model", an application model of vehicle-cloud collaboration, not traditional single-vehicle intelligence. When the FSD decision is inconsistent with the driver's decision, the data is sent back to the cloud for FSD training, and the training results are optimized and deployed OTA on the vehicle side, so that the driving level continues to approach the human level.

In Li Keqiang's view, there are inherent drawbacks in the training data of a single car company's large model. First, the data volume is limited, and it is impossible to achieve the massiveness of training data; second, the data types are incomplete, and the completeness of training data cannot be guaranteed. The vehicle-road-cloud integrated system integrates the data acquisition mode of vehicle autonomous reporting and roadside perception unified collection, with a wider coverage, more complex traffic scenarios, and more complete data types, which has natural advantages.

As artificial intelligence and the new generation of information and communication technology accelerate the transformation of the automobile industry, my country has taken the lead in proposing an innovative development path for "vehicle-road-cloud integrated" intelligent connected vehicles. The national government has accelerated top-level design and the formulation of policies and regulations. Testing and demonstration in cities across the country have been vigorously carried out. All sectors of the industry have reached a consensus on the development path, pointing out a new direction for the development of my country's intelligent connected vehicles.

Image source: Neusoft Reach

Neusoft Rich believes that the autonomous driving industry is rapidly spreading from L2 to L3 levels, and market competition is mainly centered around sensor configuration and chip computing power, with performance and hardware costs facing bottlenecks. In this context, car companies have become innovative development paths to achieve intelligent vehicle upgrades in the future by combining cloud-based large models and edge computing communication foundations, and relying on road-side perception and real-time dynamic data analysis. Facing the continuous development of future smart cars, Neusoft Rich's car-cloud integrated full-stack products enable car companies to innovate intelligently and efficiently, and continuously iterate through the construction of vehicle-side product technology capabilities and the deployment of cloud platform capabilities, while providing forward-looking technical support for the large-scale application of intelligent connected vehicles.

How to achieve commercialization of “Car-Road-Cloud”?

"Intelligent connected vehicles are a high-tech product, and the industrialization of a high-tech product should meet two conditions: a technical closed loop and a commercial closed loop. If there is a technical closed loop, there may not be a commercial closed loop, but if there is no technical closed loop, there will definitely be no commercial closed loop," said Li Keqiang.

In Li Keqiang's view, there are two main reasons why networked intelligent driving has not yet formed a commercial closed loop. First, the current vehicle-road-cloud integration research and development and demonstration are still in the early stages, the system is mainly based on single-vehicle intelligence, and the data of car companies has not been connected and cannot be connected; second, the vast majority of current vehicle-road-cloud integration systems are still chimney-type architectures, and layered decoupling and cross-domain sharing have not been achieved. He hopes that these two problems can be solved through demonstration applications.

As a vehicle-road-cloud integrated autonomous driving company, Mushroom Autolink has had intelligent connected vehicle-road-cloud integrated projects in more than a dozen provinces and cities across the country in the past two years. Mushroom Autolink CTO Guo Xingrong told China News Weekly that there are many reasons for the problems encountered during the implementation of vehicle-road-cloud.

Image source: China Electric Vehicle 100

The first is everyone's positioning and judgment on this matter. Do we regard the roadside infrastructure and infrastructure serving intelligent networking as a public infrastructure, or do we build it as an asset that can be monetized?

In addition, should we explore the integrated vehicle-road project such as intelligent networking as a demonstration project or treat it as a large-scale application?

"In the early stages, we did a lot of small-scale construction, such as building a road of 10 kilometers or 5 kilometers, or demarcating an area of ​​10 square kilometers or 20 square kilometers. Although small-scale construction is conducive to controlling investment costs, from now on, small-scale demonstrations can no longer meet the needs of car companies and users, as well as the future development needs of the industry." Guo Xingrong said, "The general requirement of car companies is that they can be driven across the country like NOA's cars. If they are only given an area of ​​5 square kilometers or 10 square kilometers, it will not bring much benefit to users, so they are not very interested in doing this."

"The same is true for industry development. For this industry to develop rapidly, it must have economies of scale. After scale-up, on the one hand, costs can be reduced rapidly, and on the other hand, the entire industry will be driven forward." Guo Xingrong said.

Third, China has been good at the construction of traditional infrastructure in the past, such as the development and construction of roads, bridges, and real estate. On this basis, new business or technology models are superimposed, which requires new technology operators to operate. Some people now call them "new quality operators". On the one hand, "new quality operators" need to plan the selection and layout of roadside traffic facilities in the entire city, build cloud platforms, and empower autonomous driving and intelligent networked vehicles through road and cloud data. In addition, various autonomous driving vehicles must be deployed and operated locally, so the participation of many high-tech companies is needed. In addition, there needs to be such an operator who has experience in roadside, cloud, and car, and can truly integrate the data of the three terminals of car, road, and cloud. "Only construction without operation will not produce the final effect." Guo Xingrong said.

"The integrated development of vehicle, road and cloud is in line with the trend of technology and industry development. We should maintain strategic focus and continue to promote the implementation of technology." Li Keqiang said that we should look at problems objectively, fully understand the current problems that need to be solved urgently, and the solutions to the problems that arise, form high-quality solutions, and promote them in a systematic manner. "We should form a consensus on the development path, seize the development window period, continue to iterate investment, give play to the advantages of the new national system, further increase investment through overall coordination, solve problems, and achieve large-scale industrialization goals as soon as possible." Li Keqiang said.

Author: Liu Shanshan