Baidu robotaxis grounded: China’s traffic chaos exposes real-world limits

Baidu’s fleet of more than 500 autonomous vehicles in Wuhan stopped almost at the same time and turned the traffic lanes into a blockade. That matters because it shows that a robotaxi system is not just a bunch of cars; it is a shared logic layer that can fail all at once. TechRadar described the incident from the ground, while Apollo Auto remains the core platform behind Baidu’s autonomy push.

This was not a story about a single bad sensor. It was a system-wide failure, which is exactly why centralized autonomy is so brittle when it has to degrade safely. In robotaxi terms, scaling is not the same thing as safety, and Wuhan proved that in the least polite way possible.

Wuhan should have been a showcase for progress, but instead it became a reminder that real traffic does not behave like a test track. Once the whole system freezes, the demo is over and the operational bill begins.

Baidu’s own financial reporting makes it clear that Apollo Go is no toy: it is a serious service with millions of rides and expansion plans. That means every failure has reputational and commercial weight, because the fleet is no longer a pilot but infrastructure. That is precisely why this incident matters: scale and resilience are not synonyms.

The biggest weakness of fleets like this is not only the algorithm, but the lack of a truly graceful failure mode. When a vehicle cannot exit a bad state safely, autonomy turns into a very thin layer of luck.

That is a lesson for both regulators and companies: a large fleet is not impressive if it cannot fail safely.

In other words, Baidu’s problem is not speed. The problem is that the whole system stalls together. And when that happens, a robotaxi becomes traffic clutter instead of transport.

The industry takeaway is simple: autonomy needs real degradation, not just a stop button. If a fleet has to shut down because one layer of logic fails, the operating model is still too fragile for public roads.