Xiaomi’s Automobile Business “Compulsory Courses” Undergo a Major Test

Xiaomi Auto faces a “major test.”

On September 19th, Xiaomi Auto Technology Co., Ltd. filed a recall plan with the State Administration for Market Regulation in accordance with the requirements of the “Regulations on the Recall of Defective Automobile Products” and the “Measures for the Implementation of the Regulations on the Recall of Defective Automobile Products.” It was decided that starting from today, a partial recall of the standard – version SU7 electric vehicles produced from February 6, 2024, to August 30, 2025, will be carried out, totaling 116,887 vehicles.

Recall number S2025M0149I: It involves models XMA7000MBEVR2 and XMA7000MBEVR5, totaling 98,462 vehicles. Recall number S2025M0150I: It involves model BJ7000MBEVR2, totaling 18,425 vehicles.

According to the recall information released by the State Administration for Market Regulation, the reason for this recall is that “in certain situations when the L2 high – speed pilot assist driving function of the vehicle is enabled, the recognition, early warning, or handling of extremely special scenarios may be insufficient. If the driver does not intervene in time, the risk of collision may increase, posing a safety hazard.”

This recall does not require the vehicles to be sent back to the factory for repair. Through OTA technology, software updates will be remotely pushed, and car owners do not need to go to the store. Xiaomi officials said that this software upgrade will optimize the speed control strategy of the assisted driving system, add a dynamic vehicle speed function, and improve the continuity and reliability of the high – speed pilot assist function.

It is worth noting that according to the national standard “Rules for Automobile Product Recall Numbers and Their Application,” the code S in the recall number indicates a safety defect, and the code I represents a recall affected by an investigation. (Interpretation of the national standard GB/T 39061 – 2020 “Rules for Automobile Product Recall Numbers and Their Application”) This coding rule shows that this Xiaomi Auto recall is a passive recall under the influence of regulatory investigations, rather than a software upgrade and version update voluntarily initiated by the enterprise.

Of course, regardless of the method, the real meaning of a recall is upgrading and progress. It can be said that this is a “compulsory course” for an automobile enterprise. For reference, Tesla, as a pioneer in OTA recalls, handled multiple recalls in this way in 2024, involving up to 3.39 million vehicles, almost all related to the “misuse risk” of automatic assisted steering.

01

The official notice did not specify what accident the “investigation” in the code I refers to.

Looking back at the news, on March 29, 2025, a Xiaomi SU7 with the NOA intelligent assisted driving function enabled had a serious collision and caught fire on the Chiqi section of the Deshang Expressway in Tongling, Anhui, resulting in the unfortunate death of three people.

According to Xiaomi’s official statement, “Before the accident, the vehicle was in the NOA intelligent assisted driving state and was continuously driving at a speed of 116 km/h. Due to road construction on the accident section, the lane was closed with roadblocks and the traffic was diverted to the oncoming lane. After the vehicle detected the obstacle, it issued a warning and started to decelerate. Then the driver took over the vehicle and switched to manual driving, continuously decelerating and steering the vehicle. Subsequently, the vehicle collided with a cement pile of the isolation belt. The last confirmed speed of the vehicle before the collision was about 97 km/h.”

This means that before the accident, the vehicle was continuously driving at a speed of 116 km/h in the NOA intelligent assisted driving state. After detecting the roadblocks due to construction and lane diversion, it issued a warning and started to decelerate. After the driver took over, the vehicle continued to decelerate from 116 km/h to 97 km/h. During this process, the first second was in the NOA assisted driving state, and the second second was in the manual driving state.

The final investigation result of the accident has not been released yet, but this accident undoubtedly exposed the current technical limitations of the L2 – level assisted driving system in extreme scenarios: insufficient recognition ability for complex road conditions, too short early warning time, and insufficient time window for human – machine handover.

Two days before the release of the recall plan, on the 17th, the Ministry of Industry and Information Technology publicly solicited opinions on the mandatory national standard of “Safety Requirements for Combined Driving Assistance Systems of Intelligent and Connected Vehicles” (Standard Interpretation | Mandatory National Standard (Draft for Comment) of “Safety Requirements for Combined Driving Assistance Systems of Intelligent and Connected Vehicles”) (hereinafter referred to as the “new national standard”). The standard sets up all – around safety technical requirements such as human – machine interaction, functional safety and expected functional safety, information safety, and data recording for different functions such as single – lane, multi – lane, and pilot assistance. Among them, the standard constructs test scenarios for traffic environments such as road intersections, construction areas, roundabouts, and tunnels, and also adds tests on the detection and response capabilities for targets such as pedal – powered two – wheeled motorcycles, temporary obstacles, and overturned vehicles.

The new national standard puts forward new requirements for the combined driving assistance system test. Especially in Sections 7.4.10.4 and 7.5.8, the “hand – off test” and the “detection and response ability test for construction areas in Class B road environments” almost replicate the scene of the “March 29th fire accident” on a 1:1 scale.

Judging from the chronological order, this accident is very likely the origin of the code I in the recall number.

In fact, this is not Xiaomi Auto’s first recall. On January 24, 2025, the State Administration for Market Regulation issued an announcement that Xiaomi Auto filed a recall plan. It was decided that starting from today, a partial recall of the standard – version SU7 electric vehicles produced from February 6, 2024, to November 26, 2024, will be carried out, totaling 30,900 vehicles.

At that time, Xiaomi Auto said that “in certain situations when the L2 high – speed pilot assist driving function of some vehicles within the scope of this recall is enabled, the recognition, early warning, or handling of extremely special scenarios may be insufficient. If the driver does not intervene in time, the risk of collision may increase, posing a safety hazard.”

Both recalls are due to the L2 high – speed pilot assist driving function.

02

Xiaomi Auto’s recall and the public solicitation of opinions on the new national standard point directly to the insufficient safety of the L2 high – speed pilot assist driving function, which precisely touches on a core pain point in the current intelligent vehicle industry – the huge gap between the excessive marketing of assisted driving functions and the technological reality.

In the past few years, the statement “drivers can ignore the vehicle when using assisted driving on the highway” has almost become a standard selling point for new – energy vehicle enterprises. Xiaomi emphasized its NOA (Navigate on Autopilot) intelligent navigation assisted driving function in the marketing of the SU7, claiming that it can provide a “quasi – autonomous driving” experience on the highway.

Similar statements also frequently appear in the marketing of other vehicle enterprises.

On May 26 this year, the “Yu Chengdong sleeping while driving” incident was a typical example. Yu Chengdong, an executive director of Huawei, was photographed looking down at his phone for nearly 20 seconds without holding the steering wheel while driving a Wenjie M8, which caused widespread controversy. Although Yu Chengdong later clarified that he was looking at his phone while using intelligent driving instead of sleeping, this incident exposed the ambiguity in the industry’s understanding of the safety boundaries of the L2 – level assisted driving function.

What’s more, although traffic management departments have repeatedly reminded that drivers still need to monitor the vehicle at all times and be ready to take over the vehicle when using the L2 – level system, before the implementation of the new national standard, regulatory agencies had not clearly defined the concept of “high – level intelligent driving.” As a result, many vehicle enterprises defined relevant concepts on their own, causing confusion among consumers and posing great potential risks.

The “March 29th fire accident” has once again sounded the alarm for the safety of L2 – level driving, and also made the road administration department add more reminders.

Safety reminder for L2 assisted driving

After all, from January to July this year, the sales volume of new passenger cars equipped with combined driving assistance systems in China was 7.7599 million, a year – on – year increase of 21.31%. The penetration rate was 62.58%, an increase of 6.5 percentage points compared with the previous year.

Facing the increasing adoption of combined driving assistance systems in electric vehicles, the new national standard gives clear requirements: When the vehicle speed is greater than 10 km/h, the system should issue a HOR at the latest 5 s after the driver’s hands leave the steering wheel. Only when the system can detect that the driver’s eye gaze direction does not deviate from the driving – related area is the system allowed to issue a HOR at the latest 10 s after the driver’s hands leave the steering wheel. If the driver continues to keep their hands off the steering wheel, the system should upgrade the HOR at the latest 10 s after issuing the HOR.

The upcoming implementation of the new national standard undoubtedly means that the L2 – level assisted driving will have a clear national standard to follow, which will help standardize the market order and prevent safety risks.

In April this year, at the promotion meeting on the access of intelligent and connected vehicle products and the management of software online upgrades, the Ministry of Industry and Information Technology emphasized that automobile production enterprises should fully carry out combined driving assistance tests and verifications, clarify the system function boundaries and safety response measures, not conduct exaggerated and false publicity, strictly fulfill the obligation of notification, and earnestly assume the main responsibility for production consistency and quality safety, so as to effectively improve the safety level of intelligent and connected vehicle products.

In addition, the “Notice on Strengthening the Recall, Production Consistency Supervision and Management, and Standardizing the Publicity of Intelligent and Connected New – Energy Vehicle Products (Draft for Comment)” issued in August this year also clearly requires vehicle enterprises to provide consumers with information on the driving automation level, system capabilities, and system boundaries of intelligent and connected new – energy vehicles truthfully and comprehensively, and not to make false or exaggerated publicity about system capabilities or misleading publicity, so as to ensure that consumers can correctly understand and drive intelligent and connected new – energy vehicles. In the naming and marketing publicity of combined driving assistance systems or functions, enterprises shall not imply to consumers that they can be regarded as autonomous driving systems or have functions that they actually do not have, so as to prevent drivers from abusing them.

These requirements and measures will establish a tighter safety net, effectively preventing vehicle enterprises from being perfunctory in safety issues and guiding the high – quality development of the industry.

03

To be honest, all vehicle enterprises know that human lives are at stake and that they need to build good and safe vehicles. However, the technological leap from the current L2 – level assisted driving to L3 – level conditional autonomous driving may be much more difficult than the industry generally expects.

From the perspective of computing power requirements, L2 – level tasks such as automatic parking and urban NOA require computing power support of dozens to hundreds of TOPS, while L3 – level and above require breaking through the 1000 – TOPS threshold to meet the real – time reasoning requirements of end – to – end models. This is not only an increase in order of magnitude but also involves a fundamental change in the algorithm architecture.

Before the accuracy and safety reach new standards, vehicle enterprises should not boast about the high – speed experience and exaggerate the high – speed pilot assist driving function at the same time.

Xiaomi Auto implemented the recall through OTA and applied for filing, indicating that the safety defect is not at the hardware level or can be solved at the software level.

In the era of electric vehicles, in addition to solving safety defect problems, OTA has also become a way for vehicle enterprises to upgrade in – vehicle systems. From a technical perspective, the intelligent driving industry is currently undergoing a new round of technological innovation. Many vehicle enterprises are vigorously promoting the vision – language – action route centered on the VLA model, aiming to solve problems such as the interpretability of end – to – end models and long – tail scenarios. All these updates in model architecture and functions need to be carried out through OTA.

Although OTA seems similar to the “online update” of mobile phone systems in form, the important difference is that vehicles are used for a longer time than mobile phones. If there is a bug in the mobile phone system upgrade, at most it will make the system freeze or the program crash, while the intelligent driving system is related to the lives of drivers and passengers.

This characteristic of long – term maintenance and being life – critical determines that in the intelligent era, vehicle delivery is no longer the end but the starting point of service.

Whether it is a recall or a normal OTA upgrade, it is part of the service provided by vehicle enterprises. Taking Tesla as an example again, frequent recalls have not had a serious impact on Tesla’s brand image. A responsible attitude and sincere and proper problem – solving are manifestations of safety responsibility. As a new entrant in the automotive industry, Xiaomi Auto will inevitably encounter various problems on its growth path, which is a compulsory course.

With the upcoming implementation of the new national standard, for all vehicle enterprises, ensuring safety and innovating to improve technological strength means a fundamental shift in the focus of competition from marketing to building real strength. At the technical level, in addition to traditional product design and manufacturing capabilities, cloud service capabilities, software development and iteration capabilities, data processing and analysis capabilities, etc., will become new core competitiveness.

In a sense, the ultimate battle in the future automotive industry will not only be in the factory but also in the cloud.

Xiaomi Auto’s recall this time is not only a repair of technical problems but also a collective reflection of the entire intelligent vehicle industry. Preventing safety hazards and carrying out recalls are compulsory courses for all vehicle enterprises, and they must achieve good results.

From a positive perspective, Xiaomi’s recall this time marks that the industry has begun to feel the new weight of the word “recall”: while pursuing technological innovation and market competition, safety should always be an uncompromisable bottom line.

Xiaomi Auto’s two recalls are also a test paper that the entire industry has to face: safety is the bottom line. If vehicle enterprises are rigorous and serious in addressing safety issues, recalls will not become a burden and may even encourage them to improve technology and take a more responsible attitude. On the contrary, if they are perfunctory because OTA recalls seem easier, they will be under the fire of public opinion and regulatory scrutiny.

The era of strong supervision of intelligent vehicles has begun, and preventing safety hazards and carrying out recalls will become the new normal in this era. It is hoped that Xiaomi Auto and all vehicles can get a perfect score.

This article is from the WeChat official account “Most Talk FunTalk” (ID: iFuntalker), author: Lin Shu, editor: Liu Yuxiang. It is published by 36Kr with authorization.