Road Tests Uncover the Reality of China’s Smart Driving Hype

On July 23, the Chinese automotive media outlet Dongchedi, in partnership with the state-run broadcaster CCTV, released the results of an unprecedented real-world test of 36 advanced driver-assistance systems (ADAS) found in major electric vehicles sold in China.

Dongchedi is a leading Chinese automotive media platform whose name means “car expert” or “one who truly understands cars.”

The evaluation aimed to assess how these vehicles perform in complex and potentially dangerous driving scenarios that simulate real-life conditions.

The outcome was shocking: not a single vehicle passed all the scenarios. The overall performance of domestic Chinese brands fell far short of their own advertising, which often claims “high-level intelligent driving” capabilities comparable to Level 3 or even Level 4 autonomy. In reality, most systems struggled to complete basic safety maneuvers.

Tesla emerged as the only manufacturer with a strong performance. Its Model 3 and Model X achieved a near-90 percent success rate, standing far above competitors.

In contrast, domestic brands such as Huawei’s Aito, Xiaomi’s SU7, and BYD’s Han performed poorly, with several vehicles failing in nearly all key test categories.

Among the highway trials were six especially challenging events, such as the “disappearing lead car” scenario, where the vehicle must respond when the car in front suddenly veers away, and a simulation of a wild boar crossing the road. In city conditions, test scenarios included children darting into traffic and merging into roundabouts.

In highway testing, only 24 percent of the vehicles managed to complete scenarios successfully. Seventy percent crashed in the disappearing lead car test, while 53 percent failed to avoid nighttime construction trucks. The wild boar crossing scenario had a pass rate of just 4.5 percent, with most vehicles either ignoring the animal or misidentifying it as a shadow or plastic bag, resulting in failure to brake.

Urban scenarios fared only slightly better, with a pass rate of 44.2 percent. In one of the most critical cases—when a child runs into the street—42 percent of the tested vehicles did not react in time.

For instance, the Aito M9 failed to avoid an obstacle in a construction zone because its AEB and NOA systems conflicted with each other, demonstrating that even high-end hardware configurations—including LiDAR, millimeter-wave radar, and multiple cameras—are ineffective without robust system fusion and software maturity.

In sharp contrast, Tesla’s approach relies entirely on a vision-based system, enhanced by extensive global driving data and training in virtual environments using its proprietary simulation platform, the “World Simulator.”

Tesla is legally prohibited from collecting local driving data in China due to national data sovereignty laws.

Perhaps most alarming was the widespread failure of AEB systems in life-threatening situations. In many cases, the emergency braking system failed to activate at all. According to the test data, vehicles using ADAS had an accident “survival rate” of just 17 percent—dramatically lower than the levels promised by manufacturers who advertise near Level 3 or Level 4 autonomous driving.

In recent years, domestic EV brands have competed to position intelligent driving as a defining advantage.

Huawei’s Richard Yu has declared its ADS 2.0 system “far ahead of Tesla,” even claiming the M9 is “almost impossible to crash.” Xiaomi CEO Lei Jun stated the SU7’s AEB could stop safely at 84 miles (135 km) per hour and that driver takeovers would be needed less than once every 186 miles (300 km). Even BYD’s budget Seagull, priced under 70,000 yuan (about $9,751), is marketed as having a “God’s Eye” intelligent system.

These slogans have painted a picture of effortless, near-autonomous driving—one that the test results contradict.

To further shield themselves from liability, some automakers appear to have implemented “last-second exit” mechanisms, in which the ADAS system disengages just before a collision and hands control back to the driver. This strategy enables companies to argue that any accidents occurring under manual operation are not their responsibility. Following crashes, manufacturers routinely blame user error, such as “hands-off violations” or “inadequate supervision,” even when system failures clearly contributed.

Several companies have also used patriotic messaging to boost brand appeal. Phrases like “domestic tech surpasses Tesla” and “autonomous innovation leads the world” have been used to rally nationalist sentiment and distract from technical shortcomings.

However, such emotional manipulation only deepens public misinformation and may ultimately backfire if trust is broken.

Tesla’s success stems from unified system architecture, global data networks, and rigorous simulation. Its vision-only approach avoids the internal conflicts that plague multi-sensor systems, while its World Simulator enables learning from rare or dangerous edge cases. Cost-wise, its camera-based solution is significantly cheaper than the expensive LiDAR-based systems favored by Chinese brands, allowing Tesla to redirect resources into algorithm development.

Chinese brands, by contrast, often rely on hardware-heavy builds with insufficient software support. This results in poor coordination among sensors, weak generalization in AI behavior, and high system costs that do not translate into better safety. Even when priced at a 40,000 yuan (about $5,572) premium—as in Xiaomi’s SU7—the ADAS package failed to deliver real-world reliability.

According to the international SAE standard, all tested vehicles, including Tesla’s, still fall under Level 2 autonomy, which requires full-time human supervision. Tesla’s systems, though more advanced, are not without fault, having experienced incidents overseas involving phantom braking, erratic steering, and traffic light misjudgment. This reaffirms that even the global leader in ADAS is still far from achieving full self-driving capability.

At the policy level, China’s regulatory framework for ADAS remains vague and inconsistent.

Terms like “L2++” and “high-level autonomy” are not standardized and are often used misleadingly in marketing. In the event of accidents, companies typically point to disclaimers absolving them of responsibility, making consumer redress difficult.

Moreover, beneath the glossy marketing facade lies an industry plagued by short-sightedness, a desire for quick success, lax oversight, and a skewed public perception.

Without stricter regulation, transparent communication, and a shift from hype to solid engineering principles, the industry risks not only public backlash but also long-term damage to its global credibility.

The path toward safe and reliable autonomous driving must be built on real-world data, robust algorithms, independent testing, and accountability—not just press conferences and patriotic slogans.