Xiaomi SU7 Car Accident Claims 3 Lives, so-called “Smart Driving” Under Question

In recent days, the Xiaomi SU7 car accident has sparked continuous public attention on topics such as smart driving safety, electric vehicle fires, and door unlocking, which are the most concerned issues for consumers.

Xiaomi’s official response to the accident has raised more doubts about Xiaomi’s smart driving system within the industry. Some professionals pointed out that the current Chinese electric vehicle industry has been engaging in excessive promotion of intelligent driving, which poses hidden dangers to traffic safety.

On the night of March 29, a standard version of Xiaomi SU7 on the Deshang Expressway in Tongling, Anhui, crashed into the guardrail and caught fire, resulting in the death of three people inside the car.

An anonymous industry analyst mentioned to Jiemian News that the accident has triggered widespread coverage due to the involvement of electric vehicle fires, smart driving safety, and whether the car doors can be unlocked, all of which are the most concerning topics for consumers. This incident is the result of a combination of these factors as well as the continuous public attention on Lei Jun and Xiaomi’s auto business in the past year.

Three days after the accident, on April 1st, Xiaomi Group made its first public response but failed to address concerns regarding battery fires, door unlocking, and the status of AEB (Autonomous Emergency Braking) and NOA (Navigate on Autopilot) systems. It should be noted that the NOA system and AEB system are not interrelated but are two independent functional systems.

According to the information submitted by Xiaomi to the police, the sequence of events leading to the collision unfolded as follows:

– March 29, 10:27:17 PM: NOA activated, vehicle speed 116 km/h
– March 29, 10:28:17 PM: Slight distraction warning
– March 29, 10:36:48 PM: NOA issued a hand-off warning to hold the steering wheel
– March 29, 10:44:24 PM: NOA issued a risk warning to “watch out for obstacles ahead,” requested deceleration, and started slowing down
– March 29, 10:44:25 PM: NOA was taken over, entered manual mode, steering wheel turned 22.0625 degrees to the left, brake pedal at 31% openness
– March 29, 10:44:26 PM: Steering wheel turned 1.0625 degrees to the right, brake pedal at 38% openness
– March 29, between 10:44:26-28 PM: The vehicle collided with the cement guardrail at a speed of approximately 97 km/h.

Based on Xiaomi’s response, from the NOA system issuing a deceleration request to the user taking control until the collision occurred, there was only about a 2-second window.

The mother of one of the victims, Xiao Luo, told Pengpai News that her child was driving the smart car at the time of the accident. She questioned who could react to such a situation within 2 seconds, highlighting the difficulties faced in such critical moments.

Investor in intelligent driving field, Huang Song, told First Financial, “According to the data they (Xiaomi) released, the time they left for (braking) was too short.”

Huxiu’s article “Xiaomi Accident, The Blame on Smart Driving” noted that at around 116 km/h, with only a 2-second reaction window, a driver would need about 35m/s reaction speed, which is astronomically challenging for any ordinary driver.

Jiemian News referenced reports from the US National Highway Traffic Safety Administration and the German Allgemeiner Deutscher Automobil-Club, which stated that 72% of autonomous driving accidents occur within 2 seconds after system alerts, while drivers on average need 2.3 seconds to resume control. In highway scenarios, this value may extend to 2.6 seconds.

According to the Society of Automotive Engineers (SAE) definition, automobile autonomous driving is classified into five levels: L0~L5. However, the details are too technical for general users to understand. Li Xiang, CEO of Ideal Automobile, simplified it as L1=Partial Assisted Driving, L2=Assisted Driving; L3=Automated Assisted Driving; L4=Automated Driving; L5=Driverless Driving.

Currently, domestic electric vehicle manufacturers’ smart driving systems are at the L2 level, and vehicles equipped with L3 level and above autonomous driving functions, in case of accidents, will be the responsibility of the automaker.

The standard version of Xiaomi SU7 is equipped with Xiaomi’s own Xiaomi Pilot Pro system, utilizing a dual-camera pure visual smart driving solution.

Huang Song mentioned to First Financial, “Cameras can now perceive distances of about 100 to 200 meters, with most reaching 200 meters. This 200 meters can ensure that vehicles traveling at speeds below 120 km/h can brake automatically.” However, industry insiders pointed out that current vehicle smart driving technologies are still immature, which is a reality that must be faced. For instance, the sensitivity of camera vision perception significantly decreases during nighttime and adverse weather conditions.

As per Xiaomi’s official website, the standard version of Xiaomi SU7 is equipped with 1 millimeter-wave radar, 11 external cameras, and 12 ultrasonic radars. However, based on the vehicle speed of 116 km/h at the time of the accident and the 2-second reaction time, the distance from the warning issued by Xiaomi NOA to the final collision is estimated to be around 64 meters.

Questions have also been raised about Xiaomi’s smart driving algorithm. A smart driving technician told Jiemian News that at the time, Xiaomi NOA had issued a deceleration request and started slowing down, indicating that it was about to evade obstacles. “If Xiaomi NOA’s algorithm was strong enough, it might not have collided if it had continued its operation.”

Huang Song also pointed out that the core issue lies in the software algorithm. “In terms of vision, it’s about calculation and estimating distances. The core issue, in my opinion, is that their algorithm isn’t good enough. Perhaps their perception algorithm failed to detect temporary road obstacles.”

According to Xiaomi’s official website, Xiaomi SU7 standard version is equipped with the NVIDIA DRIVE Orin chip, but this Orin chip is not the mainstream high-level solution Orin-X, but a more entry-level Orin-N, with a computing power of 84 TOPS, only 1/6 of the 508 TOPS calculation power of the Orin-X chip.

An autonomous driving research developer told First Financial, AEB (Autonomous Emergency Braking) is a feature already installed on many vehicles. It operates by real-time monitoring of the vehicle’s condition and surrounding traffic environment through driver assistance systems installed on the vehicle – which can be cameras, millimeter-wave radar, LIDAR, or a combination of various sensors – to analyze and calculate, determining the appropriate braking intervention timing.

In March 2024, during the launch event of Xiaomi’s first vehicle model SU7, Lei Jun claimed that the AEB emergency braking function successfully identified a stationary faulty vehicle in daytime high-speed scenarios at 135 km/h and nighttime high-speed at 120 km/h, and stopped immediately.

In addition, Xiaomi’s auto safety test results, including three AEB-related tests – disappearing vehicle at 100 km/h, stationary faulty vehicle at 120 km/h at night, and stationary faulty vehicle at 135 km/h, all scored “challenge successful” for Xiaomi SU7.

In their response on April 1st, Xiaomi stated that in this incident, NOA had already initiated deceleration after warning about obstacles. Approximately 1 second later, the driver took over, and NOA’s functionality was disabled.

The standard version of Xiaomi SU7 includes forward collision prevention assistance features such as Collision Warning (FCW) and Emergency Brake Assist (AEB), focusing on vehicles, pedestrians, and two-wheelers, with AEB operating speeds between 8 and 135 km/h. This function, similar to industry-standard AEB features, currently does not respond to obstacles such as traffic cones, traffic barrels, stones, or animals.

Analyzing the situation, Jiemian News suggests that based on Xiaomi’s explanation, the driver took over the vehicle after receiving NOA’s warning, causing NOA to deactivate due to braking. However, both Collision Warning (FCW) and Emergency Brake Assist (AEB) sub-functions of the standard version of Xiaomi SU7 did not come into play, ultimately leading to the collision, with the vehicle speed at 97 km/h upon impact.

A smart driving technician confirmed Xiaomi’s statement, mentioning that obstacles like traffic cones, traffic barrels, and stone piles have always been challenging for AEB system development. These types of obstacles are diverse and easy to perceive on the sensing end but can cause interference in control decisions. To avoid frequent triggering of AEB due to continuously recognizing these obstacles while driving, algorithm engineers may choose to deliberately overlook such obstacles.

German broadcaster Deutsche Welle cited German magazine “Der Spiegel” online edition’s report in May 2024 on Xiaomi SU7 test drive, stating, “In terms of assisted driving, when SU7 was first introduced last year, it did not perform well. ‘Other models can now maneuver through densely packed crowds with hands off, but Xiaomi fails to function when lane markings are missing, let alone pedestrians, cyclists, or erratic lane-changers.'”

German automotive outlet AutoBild wrote, “SU7 rapidly became a hit in China – even surpassing Tesla Model 3, but its assisted driving has long faced criticisms. Tests show that once lane markings disappear, or if the traffic situation is not straightforward, the system loses functionality, unlike Tesla or BYD.”

Multiple media outlets like Pengpai News reported that the mother of a female driver who perished in the accident once advised her not to blindly trust smart driving systems and rather feel secure while driving herself. However, the girl disagreed, often arguing that smart driving is convenient and safe, citing various justifications for safety.

Numerous industry insiders interviewed by Jiemian News pointed out that the current industry’s excessive promotion of intelligent driving has set unrealistic expectations among users, creating high hopes that do not align with actual system capabilities. However, curbing this behavior at its origin is challenging as automakers rely on aggressive marketing to establish a leading technical brand image, stacking the odds in their favor in the automotive industry’s high-stakes competition.

“Zhigu Trends,” repeatedly selected for many years as the most influential financial self-media by the Hurun Research Institute, published an article titled “Is Xiaomi Bearing the Blame for the Entire Industry?” on April 2nd, indicating that the entire automobile industry shares responsibility for this serious accident. It is not an attempt to absolve Xiaomi. Over the past few years, whether they are executives of major auto companies, automotive educational bloggers, or car dealers’ salespersons, all have been hyping ‘smart driving,’ collectively creating the myth of ‘invincible smart driving,’ instilling the misconception that ‘smart driving = automatic driving = absolute safety’ in many individuals.

The article mentioned that “Jinan Times” and other media outlets discovered salespersons aggressively promoting the powerful functions of smart driving systems, claiming to achieve ‘fully automated driving’ on highways and expressways, even stating that ‘we slept in the car while test driving at high speeds.’

Bankrupt and restructured new energy automobile brand Weimar publicly claimed in September 2020 that their first new production vehicle equipped with AVP autonomous parking technology became the world’s fastest and earliest mass-produced L4 level autonomous driving technology brand.

On March 10, 2020, Changan Automobile announced the successful holding of China’s first L3 level autonomous driving mass experience, themed “Leading the Future with Technological Mass Production – L3 Level Autonomous Driving, Liberating Hands, Feet, and Eyes,” in Chongqing.

A tech professional engaged in smart driving development for domestic brands disclosed to Jiemian News that a typical smart driving test undergoes extensive software, hardware, and road testing. However, some car companies, in order to showcase advanced smart driving capabilities to users, push smart driving systems prematurely to users for data collection and iteration during driving when functional testing is not thorough enough.

On March 20th, Miao Wei, former Minister of Industry and Information Technology, at the book launch of “Overtaking: China’s Solution for Intelligent Connected Vehicles,” mentioned, “Some companies, for marketing reasons, promote the concept of ‘advanced smart driving.’ I think this is mainly for show, a flashy concept that may mislead consumers. In fact, what’s more important is to truly improve safety standards.”

Zhang Xiang, Secretary-General of the International Association of Intelligent Transport Technology, stated that most smart driving systems on the market are at the L2 level, falling under the category of assisted driving. However, during promotional activities, companies tend to exaggerate by promoting automated driving assistance functions as “L2+,” “L2.5,” or even “L2.999,” which contradict national standards, posing the risk of showmanship and misleading consumers. Due to heavy marketing efforts by automakers, some vehicle owners excessively rely on automated driving assistance functions, raising safety concerns in traffic.