Overview
This episode conducts extreme safety tests on advanced driving assistance systems of 36 mainstream car brands in a temporarily closed real highway section, using various real accident scenarios to comprehensively reveal their performance and limitations in complex traffic environments.
Testing Background and Methods
- The test site is a 15-kilometer real highway section simulating common major accident scenarios.
- Each car has two survival chances during the day and night respectively; a single mistake results in the loss of one "life".
- The test involves traditional brands, joint ventures, and new Chinese forces, covering various types of driving assistance systems.
Main Accident Scenarios and Performance
- "Disappearing Lead Vehicle" tests following distance and avoidance strategies; most models follow too closely and react sluggishly, failing to avoid in time.
- "Temporary Construction Zone" simulates construction areas; the vast majority of models fail to recognize construction signs and obstacles, unable to avoid effectively.
- "Construction Zone + Truck" combines stationary obstacles; most models show unstable perception, with low-cost models like BYD Seagull performing more conservatively.
- "Nighttime Stationary Accident Vehicle" tests perception and reaction under low visibility; only very few models can effectively avoid.
- "Sudden Lateral Vehicle Lane Change" examines lane change reactions; most models fail to decelerate quickly and evade.
- "Crossing Wild Boar" ultimate challenge; almost all models fail to form effective recognition and braking, with only a few vehicles showing some response.
Specific Brand Performance and Typical Issues
- Some high-end models (e.g., Wenjie M9, Li Auto L6) have strong hardware but obvious practical response errors.
- Some models like XPeng, Tesla, NIO adopt strategies of increasing following distance and decisive braking, performing relatively well.
- Low-cost cars like BYD Seagull show very conservative strategies, which ironically avoid some risks.
- Most manufacturers' intelligent assistance hesitates, is indecisive, or reckless at critical moments, failing to prioritize safety.
- Some models have structural shortcomings in perception and decision logic, with near-zero recognition of sudden irregular obstacles (e.g., animals).
Key Findings and Conclusions
- Advanced driving assistance systems are far from achieving "fully autonomous driving" capabilities; drivers must be ready to take over in almost all scenarios.
- Expensive hardware does not equate to outstanding safety performance; algorithms, strategies, and R&D teams' emphasis on "safety first" is more critical.
- In real accident scenarios, the vast majority of vehicles cannot sufficiently recognize obstacles or react timely, easily causing serious accidents.
- Current driving assistance systems focus more on traffic efficiency than extreme safety; blind trust in the system carries high risks.
- The safest approach remains "human-machine co-driving," with driving assistance only as a risk reduction supplement; core safety depends on the driver.
Recommendations and Reminders
- Driving assistance systems should not over-promote "safety" but emphasize their limitations and role as auxiliary tools.
- Users should understand system boundaries and decisively take over in complex situations.
- Automakers need to focus on testing and optimizing systems in complex real scenarios to improve perception and escape capabilities.
- The industry needs to establish more comprehensive and stringent testing standards to truly promote technology serving safety.
Decisions
- Emphasize safety execution rather than mere safety claims.
- Insist on human-machine co-driving; drivers must be ready to take over at any time.
Action Items
- TBD – Automaker R&D Teams: Optimize driving assistance system decision logic to improve responses to irregular obstacles and complex scenarios.
- TBD – Testing Organizations/Industry Associations: Promote the formulation of stricter industry testing standards.
- TBD – Users/Drivers: Enhance safety awareness, use driving assistance reasonably, and actively learn product boundaries and manual takeover timing.