True Autonomous Driving Safety Can't Be Solved by Hosting a Few Symposia

[!quote]

When an industry needs regulators to personally step in and shout "stop irrational competition," it means the bubble has grown so large that even the referees can't stand watching anymore.

The signal from this MIIT symposium is actually quite clear: the autonomous driving track is overheating, hot enough to need cooling down. But the question is, can a single notice and a few meetings really resolve the two core contradictions of "irrational competition" and "safety assessment"?

In the short term, this is a necessary brake. Over the past two years, automakers have hyped up their autonomous driving capabilities to the sky in order to grab headlines like "First L3 Launch" or "First City NOA." Some manufacturers have even set the trigger rate for AEB (Automatic Emergency Braking) extremely low, just for that one "zero takeover" demo video at launch events. This kind of involution-style competition is essentially betting with consumer safety. By stepping in now, MIIT can at least serve as a deterrent—forcing those companies cutting corners in testing to behave themselves a bit more.

But in the long run, this symposium will likely devolve into a "formalist" show.

First, what are the standards for safety assessment? Currently, domestic testing and verification for autonomous driving still relies primarily on self-certification by enterprises. There are no unified third-party testing standards, no public crash datasets, and no mandatory hardware redundancy requirements. The symposium calls for "strengthening safety assessments," but how exactly should they be assessed? Who does the assessing? What are the penalties for failing? None of these core issues have been resolved. Requirements without detailed rules are just "persuasion."

Second, the essence of irrational competition is market anxiety. Why are automakers desperately burning cash and stacking specs? Because behind them is the valuation logic of the capital markets. Whoever shouts "L4 mass production" first gets the next round of funding. This anxiety isn't something you can eliminate with one meeting. As long as capital continues to chase the autonomous driving concept, companies will find various gray-area ways to bypass regulations. For example, packaging "L2+ level" as "advanced smart driving," or marketing "driver assistance" as "autonomous driving." MIIT can control the wording at launch events, but it can't control the sales pitches used by dealerships.

Third, the real safety risks aren't in the lab, they're on actual roads. Corner cases for autonomous driving are virtually endless. China's complex road conditions—pedestrians popping out suddenly, three-wheeled vehicles going against traffic, jaywalkers, animals appearing unexpectedly—these cannot be covered by simulation tests alone. And to cut costs, automakers often prioritize slashing "road test mileage" and "scenario library construction." The symposium emphasizes "identifying hidden safety risks," but if the result of the investigation is just "risks exist but cannot be immediately resolved," what's the point?

I always believe that the speed of autonomous driving deployment far exceeds the maturity of safety technology. There is a strange consensus in the industry right now: as long as the code runs, the tech passes. But true engineering implementation requires redundant design, fault tolerance, and full coverage of extreme scenarios. These hard skills cannot be solved just by throwing money at hiring a few algorithm engineers.

So, what substantive changes can this symposium bring?

I tend to see it as a signal: regulators are beginning to realize that the autonomous driving bubble threatens the health of the industry. But a signal is not action. If there are no accompanying laws and regulations, mandatory technical standards, or independent third-party testing agencies afterwards, then this symposium is just a gentle warning. Automakers accustomed to "jumping the gun" will likely continue probing the gray areas.

One final question: When an automaker claims at a launch event, "Our autonomous driving system has undergone a million kilometers of road testing," how many people know that half of those million kilometers might have been driven in closed venues, good weather, and interference-free conditions? Who exactly defines the word "safe" in autonomous driving, and who is responsible for it?

Original link: https://www.ithome.com/0/978/299.htm