Wuhan Driverless Car Glitch Sparks Safety Controversy — What Baidu's Apollo Go Outage Reveals About Autonomous Fleets
AI / AI Policy | 5 min read
On the evening of 31 March 2026, more than 100 Baidu Apollo Go robotaxis simultaneously came to a complete standstill in the middle of busy roads, elevated expressways, and fast-moving intersections across Wuhan, China — trapping passengers inside vehicles, snarling traffic for hours, and reigniting an urgent global debate about the safety and readiness of autonomous transport at scale. Wuhan's Municipal Public Security Bureau confirmed that beginning at 20:57 local time, multiple Apollo Go autonomous taxis halted across major roads and elevated ring roads due to what authorities described as a "system malfunction". No injuries were reported and all passengers were ultimately evacuated safely — but some riders were trapped for nearly two hours, with in-car SOS buttons unresponsive and customer service lines overwhelmed.
What Happened: The Scale and Nature of the Failure
Wuhan is the world's largest deployment site for Apollo Go, with more than 1,000 fully driverless vehicles operating across the city. When the malfunction struck, the robotaxis did not pull to the side of the road as a human driver would. They simply stopped — in the middle of lanes, at intersections, on elevated highways with fast-moving traffic. Hazard lights flashed, but the vehicles were immobile. Social media was quickly flooded with footage showing lines of white SUVs forming gridlocked barriers across major arteries. Some passengers were able to open doors and exit, but many hesitated due to the dangerous traffic conditions and called police for assistance. Emergency response teams and Apollo Go staff worked through the night to evacuate passengers vehicle by vehicle and tow the malfunctioning cars. Traffic order was not fully restored until the early hours of 1 April. Industry analysts suggest the failure may have been triggered by abnormalities in cloud communication or vulnerabilities in system algorithms — causing the vehicles to activate a "safe stop" mechanism. The exact cause remains under investigation, and Baidu has not published a detailed technical explanation.
The Correlated Failure Problem: A New Category of Risk
The most significant insight from the Wuhan incident is not that one autonomous vehicle failed — it is that more than 100 failed simultaneously, in exactly the same way, at exactly the same time. This is what analysts are calling correlated failure — a category of risk that is structurally different from anything that human-driven fleets face. When a human driver makes an error, it affects one vehicle. When a software bug exists in a cloud-managed autonomous fleet, it can propagate instantaneously across every vehicle connected to the same system — turning a single point of failure into a citywide transport crisis. Unlike a traditional vehicular breakdown, the robotaxi failure mode is not mechanical or localised: it is systemic, simultaneous, and scalable. Experts note that as these fleets grow, so does the potential blast radius of any system-level failure. The Wuhan outage, involving 100+ vehicles, is a fraction of the scale that could be affected as autonomous fleets expand to thousands of vehicles across dozens of cities.
The Timing: Baidu's Global Expansion at a Sensitive Moment
The incident arrives at a particularly consequential moment for Baidu. In December 2025, ride-sharing giants Uber and Lyft announced landmark partnerships with Baidu to bring Apollo Go vehicles to the UK — with pilot programmes in London scheduled to begin in the first half of 2026 using the Apollo RT6, the sixth-generation fully autonomous robotaxi equipped with eight LiDAR sensors, 12 cameras, 12 ultrasonic sensors, and one wave radar. Uber has also announced plans to deploy Apollo Go in Dubai by 2030. Apollo Go delivered 3.4 million fully driverless rides in Q4 2025 alone, with the fleet having covered more than 300 million kilometres autonomously — over 190 million without a human safety driver. Footage of passengers stranded on elevated highways in Wuhan will not reassure London or Dubai regulators already applying scrutiny to the technology. The incident is not the first safety challenge for Apollo Go — an earlier robotaxi fell into a construction pit in Chongqing in August 2025, and Pony.ai's vehicle caught fire in Beijing in May — but the scale and visibility of the Wuhan outage marks a new threshold in public and regulatory attention.
Key Takeaways
- • On 31 March 2026 at 20:57 local time, 100+ Baidu Apollo Go robotaxis simultaneously halted mid-traffic across Wuhan's roads and elevated expressways — trapping passengers for up to two hours, with in-car SOS buttons unresponsive. No injuries were reported; all passengers were eventually evacuated. Wuhan authorities confirmed a "system malfunction."
- • Unlike human driver failures — which affect one vehicle — this is a case of correlated failure: a system-level fault that propagated simultaneously across every connected vehicle in the fleet, forming gridlock barriers across a major city. This is a structurally new category of transport risk that existing regulatory frameworks are not designed to handle.
- • Analysts attribute the failure to probable cloud communication abnormalities or algorithm vulnerabilities triggering a "safe stop" mechanism — which, while designed to prevent accidents, proved disastrous in real-world mixed traffic conditions where the vehicles blocked fast-moving lanes rather than pulling to safety.
- • The incident is acutely timed: Baidu has landmark partnerships with Uber and Lyft for London pilot programmes in H1 2026 (using the Apollo RT6), and Dubai deployment plans by 2030. Footage of passengers stranded on Chinese ring roads will face intense scrutiny from Western regulators assessing readiness for these markets.
- • Experts call for stronger system redundancy design, better emergency response mechanisms, and regulatory frameworks that address the unique failure modes of cloud-managed autonomous fleets — moving the industry debate from whether autonomous vehicles can drive safely most of the time to whether the systems managing them are resilient enough for real-world deployment at scale.