For L4 operators still seeking sustainable returns, such system-wide failures may prove more challenging than isolated traffic incidents.
A passenger rides in a Baidu Apollo Go robotaxi in Wuhan, China, March 28, 2026.Photo:CFP
by WANG Zhen
Passengers were left stranded for hours in Wuhan, a major central Chinese city, after a widespread outage hit Apollo Go, Baidu's robotaxi service, raising fresh concerns over the reliability of autonomous driving at scale.
The disruption occurred on the evening of March 31, when multiple robotaxis stalled across the city. Wuhan traffic police said early on April 1 that several vehicles had stopped in the middle of roads due to a suspected system failure. No injuries were reported, although videos circulating online showed some cars being rear-ended.
One passenger, identified as LU Guanggu, told Jiemian News his robotaxi came to an abrupt stop on an elevated section of the city’s Third Ring Road and was unable to move for an extended period. A journey that typically takes about 20 minutes stretched to nearly two hours.
The app continued to show the trip as active, while repeated attempts to reach customer service went unanswered. About 30 minutes later, an automated voice message in the vehicle attributed the disruption to a "network issue". Apollo Go's customer service said operations in Wuhan had resumed but declined to provide details. Baidu had not issued a public response as of publication.
Media reports suggested that at least 100 robotaxis were affected, pointing to a system-level failure rather than an isolated malfunction.
Baidu's Apollo Go robotaxi service operates in several Chinese cities including Beijing, Shanghai and Guangzhou, as well as overseas markets. Wuhan is one of its key testing grounds, where the service has moved from pilot programs to early-stage commercial operations.
ZHU Xichan, a professor at Tongji University, said the simultaneous stoppage of multiple vehicles was more likely caused by a breakdown in communication between vehicles and the backend system, rather than a failure of the autonomous driving software itself. Disruptions to communication links can hinder remote monitoring and intervention, exposing vulnerabilities in system design, he said.
The incident also exposed gaps in emergency handling. Ideally, affected vehicles should be able to pull over safely to the roadside instead of stopping in active traffic lanes, according to Zhu. Lu said the vehicle only pulled off the main carriageway after the in-car SOS system responded. Although the doors could be opened, he chose not to exit due to heavy traffic passing nearby. He said the compensation offered — a free ride, another free trip and two 20 yuan vouchers — did not adequately reflect the time lost or the distress caused.
Industry analysts said communication failures should not lead to large-scale fleet paralysis, nor should operators delay intervention once faults occur. A senior autonomous driving analyst told Jiemian News the incident highlights the need to reassess redundancy across key layers — including onboard systems, backend platforms and communication links — as well as the adequacy of contingency plans, backup systems and passenger evacuation options.
The outage has also renewed debate over the readiness of Level 4 autonomous driving (L4) — systems designed to operate without human intervention in most conditions — for real-world, large-scale deployment. ZHU said such incidents are unlikely to alter the industry’s long-term commitment to the L4 pathway, but will force a closer examination of weaknesses in backend response, incident takeover and operational handling.
Rather than an isolated vehicle failure, the simultaneous stoppage of multiple robotaxis in Wuhan exposed how the system performs under stress. Vehicles stranded in active lanes, passengers trapped for extended periods and delayed platform response point to challenges that go beyond single-vehicle capability, testing the resilience of the entire operational stack in real urban conditions.
The analyst added that the incident underscores that robotaxi commercialization cannot rely solely on technological capability. It also requires stronger system governance, clearer regulatory frameworks and more robust redundancy design, ultimately ensuring passenger safety and service reliability.
As one of Apollo Go's core operating cities, Wuhan plays a central role in its commercial rollout. Disruptions of this scale not only affect short-term operations but also touch on user trust, traffic order and broader perceptions of the maturity of robotaxi deployment. For L4 operators still seeking sustainable returns, such system-wide failures may prove more challenging than isolated traffic incidents.