HomeArticle

Behind the Xiaomi car accident: Stunning intelligent driving, a hasty two seconds

徐蔡钰2025-04-01 22:46
Noisy technical jargon and invisible risk warnings

On March 29th, a serious traffic accident occurred in Tongling City, Anhui Province. After a Xiaomi SU7 crashed into a cement guardrail, it caught fire, and the three occupants in the car unfortunately died. The accident vehicle was burned until the white body was exposed.

On April 1st, Xiaomi officially confirmed this accident and announced some driving states of the accident vehicle at that time.

According to Xiaomi's statement, before the accident, the vehicle was in the NOA intelligent assisted driving state, with a speed of 116 km/h. Due to road construction on the accident section, there were roadblocks closing the lane, and the vehicle changed to the oncoming lane. NOA is known as the pilot-assisted driving function in the industry, which can automatically control the vehicle to accelerate, decelerate, change lanes, and enter and exit ramps.

According to the driving record released by Xiaomi, after the NOA issued an obstacle warning, the driver took over the vehicle one second later. However, due to the high speed, the collision occurred within three seconds after the driver took over the vehicle. Before the collision, the final speed of the accident vehicle was still 97 km/h.

On March 30th, Xiaomi established a special working group to actively cooperate with the police in completing evidence collection and investigation work. The Transportation Bureau of Tongling City, where the accident occurred, also established a special working group and is conducting an investigation into the accident.

The protagonist of this accident, the Xiaomi SU7, is a phenomenal hit product in the Chinese automobile market in the past two years. It sold 186,000 units in one year after its launch. Currently, due to the backlog of orders, the longest delivery time is as long as one year.

Lei Jun, the founder of Xiaomi, has gradually become a top - tier figure in the automotive circle. Whether it is a press conference, an industry forum, or an offline event, almost every move of Lei Jun quickly makes it onto the hot search list. His entrepreneurial spirit has touched a large number of consumers.

Lei Jun once publicly stated that 30% of the early orders for the SU7 were from blind - order customers. They decided to purchase the car without seeing or touching it. It is the trust of consumers in Xiaomi and Lei Jun that has supported the large number of orders.

Intelligent driving is also one of the core development themes in the automotive industry in recent years. BYD, the leading new energy vehicle company, has just launched an intelligent driving equalization campaign, which has led to the follow - up of companies such as Geely and Chery. The high - speed pilot function can now be installed in cars costing less than 100,000 yuan and is used by a large number of users.

Together with new car - making companies such as Li Auto, XPeng, NIO, and Hongmeng Zhixing, which are flag - bearers of intelligent driving technology, the sales volume of cars equipped with high - speed pilot functions or above this year will directly reach the tens of millions level, accounting for as much as half of China's annual passenger car sales.

With the top - tier company Xiaomi and the surging trend of intelligent driving, this accident has already attracted national attention and will once again tighten the string of driving safety in the hearts of car companies and consumers.

Reconstruction of the accident scene: A critical two seconds

From the driving log released by Xiaomi, it can be seen that before the accident, the high - speed NOA function of this Xiaomi SU7 was turned on, and the vehicle speed was 116 km/h.

The system once issued a mild distraction alarm to the driver and a warning for taking hands off the steering wheel. After such a steering wheel hands - off alarm, the driver needs to put their hands back on the steering wheel for the alarm to disappear.

After eight minutes of normal driving, the NOA suddenly issued a reminder of "Please note that there is an obstacle ahead" and started to decelerate.

Driving log of the accident vehicle released by Xiaomi

From the time the system issued the obstacle reminder (22:44:24) to the time the vehicle collided (22:44:26 - 28), there was only a two - second interval.

Within these two seconds, the driver quickly took over the vehicle. At the time of takeover, the vehicle speed was still above 100 km/h.

The driver turned the steering wheel 22 degrees to the left and the brake pedal opening was 31%. This seems to be a hasty response. Because when stepping on the brake deeply, the pedal opening is generally around 60%. An excessive steering angle and insufficient braking force cannot stop the vehicle quickly and may easily cause the vehicle to lose control and hit surrounding obstacles.

An R & D personnel from a car company analyzed to 36Kr that against the background of a speed of 116 km/h, a 22 - degree turn of the steering wheel corresponds to a huge wheel steering angle. And the standard version of this Xiaomi SU7 is a rear - wheel - drive model. "Making a sharp turn at high speed is a very dangerous behavior, and the vehicle is likely to lose control and start skidding at that time."

The accident vehicle (source: Internet)

Judging from the picture of the accident vehicle, the vehicle hit the cement guardrail in front of the passenger seat.

An industry insider analyzed to 36Kr that at the time of the accident, the cement guardrail was located on the center line between two lanes. While the driver was changing lanes to the left, the vehicle failed to avoid in time, and the right - front side of the vehicle hit the cement guardrail. "This is an offset collision scenario, which is now also a regular test item for car companies."

Xiaomi's test at the China Insurance Automotive Safety Index (C-IASI)

In September 2024, the C - IASI released the collision test results of the Xiaomi SU7, which included the 25% offset frontal collision test scenario on the passenger side (as shown in the above picture). The test results showed that in the case of an offset collision at a speed of 64 km/h, the passengers in the Xiaomi car were not injured and received the highest - grade G rating.

Regarding the 25% and 40% side - frontal collision tests, the national standard requires a speed of 56 km/h. "In this accident, the vehicle speed was too high. The final collision speed was actually 97 km/h," an industry insider told 36Kr. "Almost no car can withstand the impact at this speed."

After the vehicle collided with the cement guardrail, it caught fire. Judging from the photos of the accident vehicle, the vehicle was burned to ashes.

A person in the battery industry told 36Kr that Xiaomi's standards for battery protection are much higher than the industry average. Even for lithium iron phosphate batteries, aerogel insulation materials are used in heat dissipation protection. However, under a high - speed impact of 97 km/h, the power battery faces a great risk of deformation and tearing.

"The standard version of the Xiaomi SU7 uses a lithium iron phosphate battery. In the case of a minor collision, a lithium iron phosphate battery will only emit smoke and will not catch fire easily. But if the internal cells are torn and the electrolyte leaks, then the fire is inevitable," an industry insider told 36Kr.

"To put it bluntly, the current thermal protection design of battery packs is basically aimed at speeds below 80 km/h. At higher speeds, the anti - collision requirements for the front cabin will increase significantly. Most car companies will not increase the overall vehicle cost for a low - probability scenario."

Generally speaking, the direct cause of this accident of the Xiaomi SU7 is that neither the vehicle's intelligent driving system nor the driver timely detected the obstacle ahead and did not leave enough time to avoid it.

Intelligent driving has become one of the important discussion points in this accident. The intelligent driving that the industry is competing to pursue and that car companies are promoting and popularizing still seems unable to handle such complex road conditions.

The noisy intelligent driving and the invisible risk warnings

"It's actually quite difficult in a scenario like at night, on a highway, with a pure - vision system, and in a road - construction situation."

Many intelligent driving industry insiders told 36Kr that the scenario encountered by the intelligent assisted driving system of the Xiaomi accident vehicle is almost a challenge that every car company's intelligent driving team will face.

The standard version of the Xiaomi SU7 involved in the accident is only equipped with an entry - level intelligent driving solution.

In terms of hardware, it includes one NVIDIA Orin N chip with an AI computing power of 84 TOPS, no lidar, and the vehicle's intelligent driving perception is achieved by one millimeter - wave radar, 11 cameras, and 12 ultrasonic radars; the software algorithm is self - developed by Xiaomi Auto and has capabilities such as high - level pilot - assisted driving.

An industry insider introduced to 36Kr Auto that in the pure - vision intelligent driving perception stage, technologies such as BEV perception (bird's - eye view) + OCC (Occupancy Network), which were popularized by Tesla in the past few years, are relatively common perception solutions in the industry.

The insider said that on the highway, the detection distance of BEV is often only about 100 meters, and in fact, the vehicle can't see very accurately beyond 60 meters. "If driving at 100 km/h, the reaction time of the intelligent driving system is basically only about 3.6 seconds."

"It's quite difficult to detect long - distance stationary obstacles. Having a lidar will definitely be better, at least it can detect obstacles earlier."

Another senior intelligent driving industry insider also said, "In the case of poor road conditions (such as poor lighting or blocked vision), it is technically difficult to completely stop the vehicle from 100 km/h; but it is relatively easy to decelerate to a relatively low speed, such as 60 km/h."

Schematic diagram of the Xiaomi SU7 accident

According to the driving log of the accident vehicle released by Xiaomi, there was only a two - second interval between the time the intelligent driving system issued the obstacle reminder and the time the collision occurred.

In addition to the cement guardrail involved in the collision, there were also traffic cones and water barriers at the construction site to guide vehicles to change lanes. The obstacle recognized by Xiaomi's NOA might be one of them.

Based on the driving log, when Xiaomi's NOA issued the reminder, the vehicle might have been 60 - 130 meters away from the obstacle.

However, in the process of assisting human driving, even if the intelligent driving system does not react in time, car companies often have another guarantee - Automatic Emergency Braking, commonly known as "AEB".

"Even if the AEB fails to stop the vehicle, within a detection range of 100 meters, the AEB is sufficient to decelerate the vehicle to 30 - 40 km/h." An industry insider told 36Kr. However, judging from the speed of 97 km/h at the time of the accident, the AEB probably did not take effect.

Previously, Xiaomi Auto had announced its own test results, claiming that it could detect a stationary faulty vehicle in the front lane at 135 km/h and successfully stop the vehicle.

Screenshot of Xiaomi's press conference

However, an industry insider told 36Kr that not only Xiaomi but also other car companies' AEBs have a rather limited range of effective scenarios. For example, when facing some static road obstacles such as water barriers, traffic cones, and guardrails, the AEB often fails to take effect.

36Kr also found in Xiaomi Auto's user manual that among the scenarios where Xiaomi's forward collision avoidance assist function cannot work properly, there are irregular - shaped obstacles such as falling rocks, lying pedestrians, water barriers, and animals.

An intelligent driving industry insider told 36Kr that although car companies have listed a long list of precautions, "it's actually just a disclaimer. And it's difficult for users to remember in which scenarios the AEB can take effect when actually driving."

In the industry, there are not a few user complaints and exposures about "vehicle scratches and collisions caused by the ineffective AEB". An industry insider told 36Kr that the pure - vision solutions of the top - tier intelligent driving car companies also receive many complaint work orders in daily life, mainly because the vehicles scratched road obstacles such as water barriers.

In other words, even though it is surrounded by various fancy new - technology terms such as end - to - end and large models, intelligent driving still makes some basic mistakes or exposes its limited ability to handle complex road conditions.

The popularization of intelligent driving has hit a safety reef.

Since the beginning of this year, car companies have launched wave after wave of intelligent driving equalization trends. Although most of the software iterations of intelligent driving functions are future plans, taking the lead in hardware to capture users' minds has become an unspoken consensus among car companies.

In the eyes of some users, the cool technology of intelligent driving caters to the trend of pioneers and young people, and those who don't use intelligent driving have become the conservatives.

According to a report from Qingdao Radio's "News Now", the girl who was driving the vehicle in the Xiaomi SU7 accident had told her mother many times that intelligent driving was "convenient and safe".

The mother once warned her daughter that the technology is not perfect yet and she shouldn't blindly trust it. It's more reassuring to drive by herself. "I said she would definitely regret it in the future; she even refuted me, saying there were various (evidences of safety)."

In fact, the evolution of intelligent driving is far from the stage where users can truly relax and rest assured.

Even an engineer who has been working in intelligent driving for many years told 36Kr Auto that he never dares to turn on intelligent driving at night because he is well aware of the current ability boundaries of the intelligent driving system.

It is even more difficult for ordinary users to distinguish the differences in capabilities brought about by different intelligent driving levels and different chip configurations of car companies, as well as the potential safety risks behind them.

An engineer told 36Kr that recently, a car company promoting intelligent driving equalization has had hundreds of failed parking demonstrations at its sales stores, which has had a negative impact on the sales conversion rate. A large number of the company's engineers