Dongchedi ist keine Eintrittskarte für Musk.

After the heavy rain in Beijing subsided in July, the program "Smart Car Testing Ground" by Dongchedi stirred up another storm on the test track: 36 popular models equipped with advanced driver - assistance systems (ADAS), ranging from the Tesla Model 3 to the AITO M9, from the Xiaomi SU7 to the BYD Han, took turns in 15 simulated high - risk accident scenarios. The test results were unexpected: the Tesla Model 3 and Model X achieved an outstanding 83.3% passing rate in high - speed scenarios.

"We spent six months refining the test scenarios, and each scenario was derived from the big data of national traffic accidents in the past three years," said Zhao Yu, an engineer in charge of the test project at Dongchedi, in an interview. The test team even visited the highway traffic police detachments in 12 provinces and cities to ensure the authenticity of the scenario design. Judging from the two released episodes, the performance of the 36 models in the "high - speed accident scenario simulation" varied significantly.

Although Dongchedi denied ranking the test results, Tesla's two models did perform better. Elon Musk quickly reposted the results on social media, captioning it with "The strength of Autopilot speaks for itself." However, the industry's reaction was far more complex. "Tesla's performance is in line with its technological accumulation, but a single test cannot define everything," pointed out Zhang Xiang, an automotive industry analyst. On the same day that Musk reposted the results, the Traffic Management Bureau of the Ministry of Public Security emphasized again through its official WeChat account: "The advanced driver - assistance function is designed to assist the driver and cannot replace the driver's operation. The driver must always stay focused."

This official stance is not without reason. According to incomplete statistics, the number of complaints about traffic accidents involving ADAS in 2024 reached 327, a 47% increase compared to the same period in 2023. "Consumers' cognitive bias towards ADAS is more dangerous than technical defects," said Li Jianlin, a member of the Expert Committee for Automobile Consumption Rights Protection of the China Consumers Association. The public disclosure of test results may exacerbate this cognitive misalignment.

More comments suggest that this test, seemingly a glorious moment for Tesla, actually exposes the hidden flaws in the intelligent driving industry: many car manufacturers use "safety redundancy" as a marketing slogan, while users are deceived by a false sense of security.

A

In the test list of "Smart Car Testing Ground", scenarios such as "sudden obstacles on the highway", "unexpected consecutive cut - ins", and "blind - spot occlusion by large vehicles" are all derived from the top 15 high - risk situations in terms of traffic accident rates. During the test, the vehicle needs to enter the preset scenario at a speed of 100 km/h, and the system must respond effectively within 1.5 seconds. This time threshold refers to the safety response standard of the Society of Automotive Engineers (SAE).

"The value of this standardized test lies in establishing a coordinate system for horizontal comparison," admitted Wu Xinzhou, the head of autonomous driving R & D at XPeng Motors, in an interview after the test. Previously, most tests in the industry were self - verification by enterprises. "You test your straight - line acceleration, and I test my cornering performance. Consumers have no way to make comparisons."

However, there are also controversies. "There is still a gap between the extremity of the test scenarios and the complexity of real - world road conditions," said Wang Liang, the director of the intelligent driving laboratory at BYD's Engineering Research Institute. The scenario of "a vehicle suddenly emerging from the emergency lane" in the test has a probability of less than 0.03% in reality due to the constraints of isolation barriers and monitoring. "Judging the system based on extremely low - probability events may mislead consumers' judgment of daily usage scenarios."

Some car manufacturers also questioned the test method. The test director of a new - energy brand revealed that their model failed in the "strong - light interference at the tunnel entrance and exit" scenario because the intensity of the simulated light source used in the test reached 8000 lumens, "which is three times the intensity of direct sunlight at noon and is almost impossible to encounter in reality." In response, Zhao Yu said: "The essence of the test is to explore the system's limit. Just like a crash test using a speed of 50 km/h, the purpose is to establish a baseline for safety redundancy."

The focus of the controversy soon shifted to the "fairness" of the test. Some netizens questioned the lack of strict control of variables, claiming that the variable settings were "full of loopholes." There were also speculations in a certain brand's owner group about whether the test vehicles were specially tuned...

This controversy highlights the deep - seated contradictions in the industry. "As ADAS moves from concept to popularization, third - party tests are necessary to break the information cocoon," said Li Keqiang, a professor at the Department of Automotive Engineering at Tsinghua University. Although Dongchedi's attempt is not perfect, "it at least makes the industry aware of the urgency of establishing a unified test standard." It is reported that the China Association of Automobile Manufacturers has initiated the formulation of the "Test Scenario Library for Intelligent Driving Systems," and the first batch of industry standards is expected to be released in 2025.

There are also relatively rational responses. "The test is like a mirror, reflecting the common technological bottlenecks of the industry in extreme scenarios," wrote Shao Mingfeng, the CBO of Voyah, on social media. His statement was overshadowed by the silence of car manufacturers - except for Tesla, almost no brand publicly claimed the test results.

In response to the doubts, tech blogger Li Nan hit the nail on the head: "In real - world accident scenarios, does anyone talk about 'fairness'?" In his view, although Dongchedi's test is not perfect, it is the first to link "laboratory data" with "real - world dangers." When an engineer from a new - energy brand privately revealed that "industry tests are often conducted under ideal lighting conditions, and the results are 30% better than in reality," this "imperfection" becomes a rare form of authenticity.

B

The problems exposed by this test are far from limited to this.

"Our car is equipped with three lidars, which is like giving the system triple insurance!" The declaration at a certain car manufacturer's press conference still rings in our ears. However, in Dongchedi's test, many models equipped with lidars failed collectively in the "nighttime backlight" scenario - it seems that the strong light overwhelmed the sensors' "clairvoyance."

This is the "safety redundancy paradox" in the era of intelligent driving: car manufacturers pile up hardware (lidars, millimeter - wave radars, high - precision maps) to create an illusion of "absolute safety," and users mistakenly believe that the system "never makes mistakes," thus relaxing their vigilance. Data shows that 47% of car owners are distracted by looking at their phones after turning on ADAS, and 87% of intelligent driving accidents recorded by the Ministry of Public Security are due to drivers completely letting go of the wheel.

"Redundant design is supposed to be a spare parachute, but some people dare to jump out of the plane without the main parachute because of it," pointed out Wang Zhenhua, a professor at the Department of Vehicle Engineering at Tsinghua University. He once disassembled a car model that advertised "L2.999" in the laboratory: its perception system collapsed after only 7 seconds in a simulated heavy - rain scenario, while the car manufacturer's manual only noted in small print to "use with caution in extreme weather."

Perhaps Elon Musk has reason to be proud: without sufficient training on Chinese road conditions, Tesla's pure - vision solution outperformed domestic cars using multi - sensor fusion, verifying the resilience of its algorithm. However, the cracks behind the "trophy" are widening - on the day after the test results were released, Tesla's stock price plummeted by 8%, and its net profit in the second quarter decreased by 20.7% year - on - year. The cold shoulder from the capital market implies a harsher truth: the "champion" in a closed - environment test may not pass the test of real - world challenges.

Not long ago, Chen Peng, a car owner in Shenzhen, posted a terrifying video on his WeChat Moments: a Tesla Model Y with Full Self - Driving (FSD) engaged suddenly accelerated in heavy rain and drove straight into a flooded area. The in - car camera was obscured by the rain like frosted glass, and the system misjudged the floodwater as a "shadow" and continued to move forward.

"The pure - vision solution is like a genius with a one - sided strength," analyzed Zhang Jing, an autonomous driving algorithm engineer. "It can get high scores in standard scenarios, but may still fail in 'out - of - syllabus' questions such as tidal lanes and left - turn lanes by borrowing other lanes, which are unique to China."

C

After the test storm, there are undercurrents in the industry. Dongchedi quietly added three new evaluation principles on its official website: Can the vehicle detect stationary obstacles at a speed of 80 km/h? Is the takeover prompt clear? What is the passing rate in rainy and foggy weather?

These simple questions point directly to the fatal blind spots in the current evaluation system.

"We should use the 'life redundancy' coefficient to replace the 'passing rate,'" proposed Chen Yu, an expert from the Intelligent Driving Sub - committee of the Society of Automotive Engineers of China. In his view, the new standard should quantify the "life - saving buffer" before the system collapses, such as whether it can issue a 1.5 - second early warning and whether it can maintain basic braking after failure.

Such practices are already taking shape: the EU's New Car Assessment Programme (NCAP) plans to include "smoothness of human - machine takeover" in its scoring system in 2026, and Tesla's "safety mode" (automatically reducing speed and turning on hazard lights when the system fails) is regarded as a model for redundant design.

The awakening of users is also driving change. Under Dongchedi's test video, a highly - upvoted comment resonated with many: "In the past, I chose a car based on parameters. Now, I only ask one question: What is the weakest road condition for this car?"

As the test results deviate from real - world safety, the industry is also reflecting on how to define safety redundancy. "The current problem is that everyone is talking about redundancy, but no one can clearly define the standard for redundancy," said Wang Yao, the secretary - general of the Intelligent Connected Vehicle Branch of the Society of Automotive Engineers of China. "Establishing a quantitative standard for safety redundancy has become an urgent task."

The lack of such a standard has directly led the industry into a "hardware race." Among the new cars launched in 2024, 63% were equipped with lidars, and the peak performance of computing chips increased by three times compared to 2022. However, tests by the Insurance Institute for Highway Safety (IIHS) in the United States showed that the correlation between hardware configuration and actual safety performance was only 0.37. "It's like installing ten airbags in a car without specifying the deployment time and coverage area. No matter how many there are, it's useless," Zhang Xiang gave an analogy.

The call for establishing a quantitative standard has received responses from multiple parties. The "Safety Regulations for Intelligent Driving Systems" being promoted by the EU clearly requires car manufacturers to disclose 23 specific parameters, such as "the recognition rate of sensors under different lighting and weather conditions" and "the takeover warning time after system failure." "Quantification is not the goal but to give consumers the right to know," said Katrin Schulz, an official from the Directorate - General for Mobility and Transport of the European Commission. After the implementation of the regulations, models that do not meet the standards will be prohibited from advertising "high - safety redundancy."

The formulation of domestic standards is also accelerating. The China New Car Assessment Program (C - NCAP) plans to add a "safety redundancy effectiveness" scoring item in its 2025 evaluation regulations, including indicators such as the anti - interference ability of sensors and the decision - making fault tolerance rate of the system. "For example, in heavy - rain weather, the effective detection distance of the radar should be maintained above 80 meters. If not, points will be deducted," introduced Liu Shiru, the head of the technical department of C - NCAP. The standard has entered the stage of soliciting opinions from car manufacturers.

For Tesla, this may pose a greater challenge. Its pure - vision approach has long been controversial for "lacking lidar redundancy." If the quantitative standard emphasizes multi - sensor fusion performance, its technical route may face pressure for adjustment.

Fortunately, a consensus is forming in the industry: the competition for safety redundancy will eventually shift from "quantity comparison" to "effectiveness comparison." "In the future, when consumers buy a car, they won't ask 'how many cameras are there,' but 'how far can it see in heavy rain,'" predicted Wang Yao, an industry observer. The establishment of a quantitative standard will force car manufacturers to shift their R & D focus from hardware stacking to algorithm optimization, "which is the real sign of the maturity of ADAS."

This discussion triggered by Dongchedi's test may ultimately point to the safety ethics of the entire industry. A netizen commented: When Elon Musk uses Tesla's test results as a technical endorsement, it is necessary to clearly recognize that the ultimate goal of ADAS is not to pass the test but to ensure user safety. As Professor Li Keqiang said: "True safety is always on the road, not on the test report."

This article is from the WeChat official account "Zimubang" (ID: wujicaijing), author: Wang Zifeng, published by 36Kr with authorization.