Autonomous Vehicles: Compliance is the Bottleneck from Launch to Road

On the evening of September 3rd, I was waiting for a cockpit interaction gray-release package in the office when I scrolled past news about Tesla's Cybercab on my phone. A friend forwarded it to me, saying driverless taxis were coming. My first reaction was to check regulations. Later, I saw that the US National Highway Traffic Safety Administration (NHTSA) launched an investigation into nearly a thousand Cybercabs, focusing on the processes and technical data Tesla relied on to determine that certain federal safety standards did not apply. The hype was immediately brought back down to earth.

The capital market's wavering on autonomous vehicles hinges on delivery. From building a cool car to long-term delivery in public roads, urban operations, accident liability, and maintenance/cleaning, there lies vehicle-grade requirements, compliance evidence, and operational backup. Launch events can push expectations high, but investigations stretch out delivery timelines.

On September 4th Eastern Time, Tesla's stock plunged nearly 6%, wiping out RMB 591 billion in market cap in a single day. Regulatory scrutiny may impact future deployment pace and expansion of operational areas.

These days, I just started using a table for breaking down smart driving promotions, having used it for less than a week. It reminds me to separate news facts, user questions, and disclosure items. The launch happened, the stock dropped, regulators investigated—these are facts. When users sit inside, they ask: Is there a steering wheel? Who do we call if something goes wrong? Who takes over if the vehicle malfunctions? Digging deeper, we need to look at disclosure items, including operational domain, takeover rate, accident statistics, remote support, maintenance stations, insurance liability, and data logging. Without these, the prettier the launch, the more it looks like a concept exhibition.

In the past month, working on smart cockpit projects, I've come to view autonomous driving as a human-machine responsibility system. No matter how smart a regular car is, the driver remains the primary entity responsible. Multiple screens, voice, and lighting in the cockpit must be designed to reduce driver distraction risks.

Cybercab is trickier because passengers might have no control authority at all. No steering wheel doesn't mean no interaction; rather, it shifts interaction from humans driving to the car explaining to humans why it is safe. When to decelerate, why change lanes, what to do when encountering pedestrians, whether clear prompts are received during system anomalies—all these need to be validated like vehicle-grade requirements.

Having worked with OTA upgrades and vehicle-grade chip toolchains recently, I increasingly feel that autonomous vehicles cannot be viewed as single-point products. If a phone OTA fails, you just restart; if a vehicle-grade OTA fails, it could affect braking, steering, perception, and power supply. Driverless taxis add scheduling, cleaning, charging, remote assistance, and insurance claims. Musk compares Robotaxi to an App and Cybercab to a pre-installed phone, with car washes and O&M centers as cloud services. The product logic seems similar. But the commercial value isn't in the phone itself; it's in whether the cloud service can run stably.

This stock reaction shows the market is pricing in the probability of realization. Previously, Robotaxi news boosted bullish enthusiasm by 8%; this time, disclosures falling short of expectations wiped out RMB 591 billion in market cap in a day. The same story gets priced differently by capital at different times. It's recalculating regulatory costs, operational costs, and expansion speed. Especially since regulators are scrutinizing the basis for exempting certain federal safety standards, which directly impacts the mass production path. Whether we can remove a steering wheel, whether we can waive certain traditional safety requirements, cannot be proven problem-free only in the lab.

This also makes me think about domestic automakers' smart driving promotions. Many launches love to say "tested" and "all scenarios." But people sitting in the car care about a different set of issues: Can it recognize construction zones? Can it identify stationary vehicles? Will it misjudge in heavy rain or backlight? Can the system alert the driver in time if they're looking at their phone? The bottom line is safety evidence. Promotions can build momentum, but regulators only certify evidence.

The real barrier for autonomous vehicles is who is responsible after an accident, how to conduct post-mortems, and whether continuous operation is possible. This exposes the industry stage from "stealing the spotlight" to "facing scrutiny": everyone has moved from watching concepts to watching compliant deliveries. Launch events are no longer enough, nor is fleet size. Next comes the competition on whether accident data feeds back into product iteration, along with functional safety, O&M networks, and responsibility boundaries keeping up.

From my perspective, this event serves as a direct reminder for cockpit product managers. Smart cockpits cannot just focus on making screens bigger and voice smoother. They must assume the role of safety explanation. When the car is driving autonomously, humans need to know what the system sees, what it plans to do, and where the risks lie. Beyond driver distraction risks, there are passenger trust risks. Trust is built through every anomaly prompt, every takeover explanation, and every accident post-mortem.

A company can tell a beautiful story but fail to provide auditable safety data, making it difficult to gain acceptance from both passengers and regulators.