Regulators Slam on the Brakes of Intelligent Driving, but Technical Dilemmas Still Need Technical Solutions | Focus Analysis

Text by Li Anqi

Edited by Li Qin

The blindly galloping intelligent driving has finally witnessed the official "braking" action.

On the evening of April 16th, an official announcement was released by the First Department of Equipment Industry of the Ministry of Industry and Information Technology. The theme of the announcement was "The First Department of Equipment Industry Holds a Promotion Meeting on the Access of Intelligent Connected Vehicle Products and the Management of Software Online Upgrades".

The announcement emphasized that automobile production enterprises need to clarify the system function boundaries and safety response measures, refrain from exaggerated and false publicity, strictly fulfill the obligation of notification, earnestly assume the main responsibility for production consistency and quality safety, and effectively improve the safety level of intelligent connected vehicle products.

Before the official release of the announcement, a screenshot summarizing the content related to the meeting had circulated in various communities in the automobile industry. The screenshot showed that 19 automobile enterprises and Huawei participated in the meeting. A person in the industry confirmed the authenticity of the content of this screenshot to 36Kr.

Judging from the official meeting notice and minutes, the publicity, function iteration and release, and function deployment of automobile enterprises' intelligent driving have been regulated and tightened to varying degrees.

For example, automobile enterprises are prohibited from using terms such as "autonomous driving", "self - driving", "intelligent driving", "high - level intelligent driving" in their intelligent driving publicity. Instead, they should describe the functions according to the automation grading standards (from Level L0 to L5) and "assisted driving".

The official also restricted the public beta tests of the so - called "thousand - person groups" and "ten - thousand - person groups" for automobile enterprises' intelligent driving, clearly stating that full verification is required and users cannot be used for testing. In addition, functions such as valet parking, one - key summoning, and remote control were also banned.

Automobile enterprises seemingly did not anticipate the "sudden" tightening of regulatory efforts by the official in advance. Some intelligent driving personnel in automobile enterprises learned the news from the news channels.

Some senior executives of automobile enterprises also told 36Kr that the regulatory authorities first conveyed the spirit of the meeting, and then everyone (automobile enterprises and intelligent driving companies) would have a discussion. That is to say, the circulated details of the notice may not necessarily be the final implementation rules. However, it is obvious that the signal of tightened control in the intelligent driving industry has been clearly sent with the fall of the regulatory hammer.

Some automobile enterprises have responded to the call of the regulatory authorities in their official publicity. For example, the latest poster promotion of Volkswagen reads "Advanced Driving Assistance System".

In fact, judging from several key events in the automobile industry this year, the standardized management of intelligent driving is almost inevitable.

Firstly, traditional automobile enterprises led by BYD have officially launched a battle for the popularization of intelligent driving. It is foreseeable that BYD will have millions of intelligent driving vehicles on the market this year. Even the Seagull model priced at around 70,000 yuan is equipped with intelligent driving functions.

At the Shanghai Auto Show, which is about to open in late April, many automobile enterprises have either joined or are preparing to join the "battlefield of intelligent driving popularization".

Secondly, half a month ago, a traffic accident involving the intelligent driving function of the Xiaomi SU7 occurred in Tongling City, Anhui Province, resulting in the tragic death of three passengers. The involved vehicle was a Xiaomi SU7 equipped with only an entry - level intelligent driving system.

Millions or even tens of millions of intelligent driving vehicles are about to be launched, but not all consumers have the ability to understand the knowledge and risks of intelligent driving. The magic of automobile enterprises trying to boost sales through "far - ahead" style intelligent driving promotion may no longer work.

When the intelligent driving functions are brought back to a unified standard, the real - life technology competition and the differences in user experience will truly emerge.

Finding the Function Boundaries of Intelligent Driving

From the circulated screenshot of the meeting minutes, the official's main intention is to clarify the current boundaries of intelligent driving, which are only in the category of Level L2 assisted human driving.

Currently, autonomous driving is divided into six levels from L0 to L5, with each level representing different degrees of automation and responsibility division. Among them, Levels L0 - L2 are driver - dominated, while Levels L3 - L5 are system - dominated.

This time, the official has more clearly regulated the publicity, function testing, OTA upgrades, scope of use, and monitoring system of automobile enterprises' current intelligent driving functions.

For example, misleading terms are prohibited in the publicity.

• Automobile enterprises are prohibited from using terms such as "autonomous driving", "self - driving", "intelligent driving", "high - level intelligent driving" in their publicity and should use "combined assisted driving" instead;
• Full Chinese names should be used as much as possible. When using English, the full name should be explained when it first appears;
• Descriptions such as "hands - off" and "eyes - off" are prohibited. "Physical disengagement" and "visual disengagement" of the driver can be used instead;
• The term "take - over" cannot be used for Level L2 assisted driving, as it is a functional term for Level L3 autonomous driving.

In terms of vehicle testing, the official has also restricted unapproved public tests, such as the "thousand - person group" and "ten - thousand - person group" tests. All tests must go through the announcement procedure. At the same time, the official also criticized applications that only fill in parameters without submitting and verifying materials.

Regarding the OTA upgrades of functions for users, the official also emphasized that functions should be fully verified before being installed in vehicles and frequent OTA upgrades should be reduced. Enterprises with frequent OTA upgrades will be subject to key reviews.

The official has also restricted the use of some functions, banning functions such as valet parking, one - key summoning, and remote control. It has also explored ways to prohibit the driver from disengaging from the driving process from a technical perspective, such as using face ID recognition in combined driving assistance to prohibit seat adjustment and reclining. Systems such as LCC (Lane Centering Control) and NOA (Navigation - assisted Driving) need to add collision - avoidance scenario tests.

Most industry insiders may not have anticipated such strong regulatory efforts by the official. However, when Tesla's FSD was launched in China, it also changed the description of the "FSD Full Self - Driving" function name on its official website to "FSD Intelligent Assisted Driving Function".

Slowing Down of Automobile Enterprises' Intelligent Driving

With the strengthening of official supervision, the work chain of automobile enterprises' intelligent driving promotion will be further calibrated and reshaped.

On the one hand, the iteration speed of automobile enterprises' intelligent driving functions will decline.

Currently, most high - level intelligent driving functions of automobile enterprises adopt the method of "pre - installing hardware + software OTA upgrades". Even if a new car is launched without intelligent driving, some automobile enterprises still use terms such as "high - level intelligent driving" and "urban NOA" in their publicity. In fact, they only provide users with entry - level intelligent driving capabilities or postpone the delivery of these functions for several months. This has become an unspoken practice among automobile enterprises.

Automobile enterprise peers are also deeply troubled by this. Last year, Li Auto quickly entered the first echelon with its intelligent driving solution of "end - to - end" + VLM (Visual Language Large Model). The engineers were in closed - door development for almost half a year, releasing a new version every month, which put great pressure on their peers.

However, the effects of automobile enterprises' OTA upgrades are not always stable. Some internal testers of intelligent driving complained to 36Kr Auto that the intelligent driving functions of automobile enterprises often experience regression after OTA upgrades. "Sometimes, just when you've figured out the capability boundaries of one intelligent driving version, the next version comes, and the effect is even worse."

After the strong intervention of supervision, the situation of frequent OTA upgrades by automobile enterprises will change. After the cancellation of the "thousand - person test group" and "ten - thousand - person test group", automobile enterprises may also need to spend more time and energy to verify the usability of each version of functions.

However, some automobile enterprises' intelligent driving may be troubled by strict OTA reviews. For example, NIO completed the development of the world model version of its intelligent driving function at the end of March this year. This is a major upgrade of NIO's intelligent driving function and a key step for NIO to return to the first echelon of intelligent driving.

However, NIO CEO Li Bin said on the social platform that the OTA version update of the world model, which was originally planned to be pushed to the public in late April, needs to wait for the official announcement due to the new regulations of the Ministry of Industry and Information Technology. "It is expected to be postponed until May."

On the other hand, the process of users using intelligent driving will also become more standardized. Previously, some users would use mineral water bottles, oranges, etc. to prop up the steering wheel to avoid the risk of hands - off in intelligent driving. Some even ignored the risks and directly fell asleep with the intelligent driving function on and the seat reclined. Some automobile enterprises also used the DMS driver monitoring system to reduce the weight of steering wheel hands - off detection in an attempt to avoid excessive interference.

However, the regulatory authorities have proposed that automobile enterprises are not allowed to suppress or turn off the steering wheel hands - off detection. They also proposed to use technologies such as face ID to identify the driver's state and prohibit seat reclining during driving, aiming for more comprehensive supervision.

The face ID recognition and the prohibition of seat adjustment and reclining proposed by the regulatory authorities also aim to achieve more comprehensive supervision through updated technologies.

A person in the industry told 36Kr Auto that this "braking" of intelligent driving actually puts all industry players on the same level, making the competition fairer. Automobile enterprises that can maintain a good user experience in intelligent driving under actual supervision will invisibly obtain a higher - level certification of intelligent driving.

Continued Technological Exploration

Generally speaking, the official mainly strengthened the management of Level L2 assisted driving this time. However, some content related to higher - level autonomous driving can perhaps be interpreted from the circulated meeting summary.

In the circulated meeting summary, the official restricted automobile enterprises from using functions such as valet parking, one - key summoning, and remote control. From the perspective of the separation between the driver and the vehicle, these functions have gone beyond the scope of assisted driving and are closer to Level L3 autonomous driving.

Starting from Level L3, vehicle control is transferred from the human driver to the autonomous driving system. However, in the state of Level L3 autonomous driving, how to determine the liability for traffic accidents is still a key and unresolved issue. Automobile enterprises have not yet reached a real industry consensus on the hardware redundancy design of Level L3 autonomous driving vehicles.

Some people from automobile enterprises and intelligent driving suppliers told 36Kr that they are currently promoting the implementation of Level L3 autonomous driving with relevant departments. Previously, Huawei, XPeng, Li Auto, etc. have all announced that they will achieve Level L3 autonomous driving this year.

It can be speculated that the functions that are currently banned are very likely to be included in the management scope of Level L3 autonomous driving in the future.

Although the regulatory authorities have pressed the "pause button" this time, their intention is not to stop the development of intelligent driving due to a single setback. More standardized management is the prerequisite for intelligent driving to transform from technology into a commodity and enter the market for long - term and healthy circulation.

However, there are still technological challenges in the intelligent driving industry. In the past two years, intelligent driving has evolved from relying on high - precision maps to using less or even no maps. The technological framework has also evolved from traditional rule - based solutions to "end - to - end" and the latest VLA (Visual Language Action Large Model), world model, etc.

With the help of the VLA large model, automobile enterprises hope that the intelligent driving system can better understand driving scenarios: identify large intersections to avoid getting lost, understand speed limit signs, distinguish complex scenarios such as tidal lanes and bus lanes, and solve the long - tail problems of autonomous driving. As a huge simulation container, the world model can also help automobile enterprises establish more realistic and rich simulation testing capabilities.

The technological bottlenecks of intelligent driving need to be solved by continuously exploring new technologies, which is both necessary and urgent. However, after curbing excessive publicity and over - exploitation of technology, the exploration of the future direction of automobile enterprises' intelligent driving may continue silently but persistently.