Baidu robotaxis freeze across Wuhan

Images

×

Image Credits:Pedro PARDO/AFP / Getty Images techcrunch.com

Kirsten Korosec techcrunch.com

A Waymo equipped with 360 degree technology including 29 cameras, five lidars (which are used for mapping and sensing) and six radar trackers. techcrunch.com

techcrunch.com

The interior of a Waymo robotaxi techcrunch.com

‹ ›

At least 100 Baidu Apollo Go robotaxis in Wuhan abruptly froze this week after what local police described as a “system failure,” according to Reuters, with some passengers stuck inside vehicles for up to two hours. Videos and social posts showed cars stopped in live traffic, including in fast lanes, and police said they were investigating the cause. Baidu did not publicly explain the outage or respond to TechCrunch’s request for comment.

A robotaxi fleet is sold as a safer version of a human driver: sensors instead of distraction, software instead of fatigue. But the Wuhan incident highlights a different failure mode—one that scales instantly. A single backend dependency, a bad update, a degraded map service, or a control-plane outage can turn hundreds of vehicles into stationary obstacles at once. In aviation and rail, safety cases assume that central systems will fail and require layered redundancy, clear “degraded mode” behavior, and rehearsed incident response. Robotaxi deployments have tended to emphasize operational expansion—more cities, more vehicles, more ride volume—while treating the remote-operations stack and fleet management systems as ordinary cloud software.

The hard question is not whether autonomous vehicles can drive; it is who carries the cost when they cannot. When a robotaxi blocks a lane or traps a passenger, the immediate burden falls on local police, tow operators, and other road users who absorb delay and risk. The passenger has no steering wheel, no ignition, and often no obvious way to exit a vehicle safely if it has stopped in traffic. Meanwhile, the accountability chain is fragmented: the operator runs the service, the OEM supplies the vehicle platform, the autonomy team ships models, and third-party vendors provide mapping, connectivity, and cloud infrastructure. Each party can plausibly argue that the failure originated elsewhere.

That fragmentation matters because it shapes incentives. If the city bears the disruption cost and the operator bears only reputational damage, the cheapest equilibrium is to expand first and litigate responsibility later. The Wuhan outage also lands as Baidu pushes into new markets: the company has said it plans to deploy more than 1,000 autonomous vehicles in Dubai over the next few years, per TechCrunch. Cross-border expansion increases the number of regulators involved, but it also increases the temptation to standardize operations on a single global control plane—the same architecture that makes a fleet-wide freeze possible.

Police in Wuhan described the incident as a “system failure.” Baidu has not said what system it was.