LabRoundupColumnNews
blog/Articles/[Shocking] 100+ Baidu Robotaxis Froze at Once in Wuhan, Trapping Passengers for 2 Hours
baidu-apollo-go-robotaxi-wuhan-mass-paralysis-cover-en

[Shocking] 100+ Baidu Robotaxis Froze at Once in Wuhan, Trapping Passengers for 2 Hours

On March 31, 2026, over 100 of Baidu's Apollo Go robotaxis simultaneously froze on Wuhan's elevated highways and roads, trapping passengers for up to 2 hours. The SOS button failed. What went wrong and what it means for autonomous driving.

News
kkm-horikawa

kkm

Backend Engineer / AWS / Django

2026.04.018 min0 views
Key takeaways

On March 31, 2026, over 100 of Baidu's Apollo Go robotaxis simultaneously froze on Wuhan's elevated highways and roads, trapping passengers for up to 2 hours. The SOS button failed. What went wrong and what it means for autonomous driving.

What Happened in Wuhan

At around 9:00 PM on March 31, 2026, more than 100 of Baidu's Apollo Go robotaxis simultaneously froze on elevated highways and main roads across Wuhan, China. Passengers were trapped inside unresponsive vehicles for up to two hours, surrounded by heavy truck traffic with no way to safely exit.

No injuries were reported. However, while the car doors could physically be opened, exiting on a busy elevated highway was too dangerous for most passengers, who chose to stay inside and call police for help.

The Wuhan Municipal Public Security Bureau's Traffic Management Division released a statement on April 1, confirming that "starting at 8:57 PM, the 122 emergency call center received multiple reports from citizens." A preliminary investigation attributed the incident to a "system failure," but Baidu has not issued an official statement as of this writing.

What Is Apollo Go

Apollo Go is Baidu's autonomous ride-hailing service, known in Chinese as "Luobo Kuaipao" (蘿蔔快跑, literally "Radish Run"). As of February 2026, it operates in 26 cities worldwide and has completed over 20 million rides. Its cumulative autonomous driving distance exceeds 300 million km, with 190 million km of fully driverless operation, making it the world's largest autonomous taxi operator by these metrics.

Wuhan, with a population of approximately 14 million, is China's eighth-largest city and one of Apollo Go's biggest deployment sites. According to a field report by Nomura Research Institute, fully driverless taxis have become an ordinary part of daily transportation in Wuhan, a reality that remains in the experimental phase in Japan.

Rides cost between 0.5 to 1.0 yuan per kilometer (approximately $0.07–$0.14), less than half the price of a human-driven taxi. This pricing advantage has fueled social controversy over the impact on taxi drivers' livelihoods. Now the mass paralysis incident has added safety concerns to the debate.

The SOS Button Did Not Work

Passenger testimonies reported by CarNewsChina paint a disturbing picture of what happened inside the stalled vehicles.

A passenger identified as Mr. Lu was trapped for nearly two hours on the Third Ring Road elevated highway, with large trucks speeding past on both sides. He reported that the in-car SOS button was "completely useless." When he tried to make calls through the backseat screen, they were automatically disconnected. He eventually called police on his personal phone, and officers along with Apollo Go staff finally arrived around 11:00 PM.

Another passenger, Ms. Zhou, waited approximately 1.5 hours inside her vehicle. The car kept displaying a warning: "Vehicle problem detected. Do not open the door." To make matters worse, she was charged the full fare despite the malfunction and forced wait.

The SOS button and remote call system, designed as the last line of defense in a driverless taxi, failed at the exact moment they were needed most. This is a more serious issue than the system failure itself.

Timeline of Events

← Swipe to navigate

How Each Party Responded

Responses to the incident varied significantly depending on the stakeholder.

Wuhan Municipal Traffic Police confirmed the reports and preliminary finding of a "system failure," but stated that a detailed investigation is ongoing.

Apollo Go Customer Service attributed the stoppage to "network issues causing driving system abnormality." However, when passenger Ms. Zhou called in, the representative claimed to be unaware of any widespread incident and said they could only investigate with a specific vehicle number.

Baidu has not released an official statement as of this writing. The company declined to comment when contacted by Bloomberg. The silence from the operator of over 100 simultaneously stalled vehicles is itself telling.

Industry insiders told CnEVPost that "the vehicles encountered unexpected circumstances that triggered a safety self-check mechanism. This is a proactive strategy by the system to ensure safety, and similar situations are not uncommon in the global autonomous driving industry."

Public Reaction

On Chinese social media platform Weibo, users with Hubei province IP addresses posted real-time reports of the incident. Posts about "Luobo Kuaipao mass paralysis" spread rapidly, reigniting debates about autonomous driving safety.

Common sentiments included: "The fares are cheap, but what do you do when it stops on an elevated highway?" and "If the SOS doesn't work, what's the point of a driverless taxi?" Others pushed back, arguing that "human taxi drivers cause accidents too — statistically, autonomous driving is safer." Public opinion remains divided.

The incident also resurfaced longstanding frustrations about Apollo Go's impact on taxi drivers' livelihoods in Wuhan, as previously reported by Sixth Tone.

The Technical Question: Is "Stopping" Really a Safety Feature?

The industry insiders' explanation that "a safety self-check mechanism was triggered" is not inherently wrong. When an autonomous vehicle detects an anomaly, stopping is a standard response known as "fail-safe" design — the principle of reverting to a safe state when something goes wrong.

But here lies the fundamental problem. The autonomous driving industry has been moving toward "fail-operational" design, which goes beyond fail-safe. Fail-safe assumes that stopping equals safety, but stopping in the middle of an elevated highway is not safe. Fail-operational systems continue operating at reduced capacity, navigating to a safe location before coming to a complete stop.

In the Wuhan incident, vehicles stopped in active traffic lanes without pulling over to the shoulder or navigating to a safe area. This suggests Apollo Go's system remained at the fail-safe stage and had not achieved fail-operational capability.

It Happened to Waymo Too

A similar situation occurred in San Francisco in December 2025. When a major power outage knocked out traffic lights, Waymo's robotaxis stalled on roads, causing traffic jams. Waymo's vehicles were designed to treat dead signals as four-way stops, but a "concentrated spike" in confirmation requests overwhelmed the remote fleet response team.

Waymo subsequently shipped a software update to improve navigation during outages and revised its emergency response protocols. In other words, Waymo followed a transparent process: problem occurred, cause disclosed, fixes implemented.

Baidu, by contrast, has disclosed neither the detailed cause nor any remedial measures. This is the critical difference between the two companies' responses.

Why the SOS System Failed

If the root cause was indeed a "network issue," it also explains why the SOS button and remote calls failed. If the emergency system depended on the same cloud connectivity as the vehicle's driving system, a network outage would take down both simultaneously.

Communication redundancy is a fundamental principle in autonomous vehicle design — maintaining multiple independent communication channels so that if the primary network fails, cellular or satellite backup ensures critical functions remain operational. At minimum, the emergency call system should operate on a communication path independent of the main system. The fact that SOS and vehicle control appeared to share the same network dependency suggests a gap in basic redundancy design.

What This Means for the Future of Autonomous Driving

Japan legalized Level 4 autonomous driving in April 2023 through a revision of the Road Traffic Act. The government has set a target of deploying autonomous services in 50 locations nationwide by fiscal year 2025, and both Toyota and Nissan plan to launch Level 4 vehicles by 2027.

Current Japanese deployments focus on low-speed services in rural areas — a far cry from the urban highways where Apollo Go operates in Wuhan. But as autonomous driving adoption accelerates, the same challenges will inevitably arise.

The Wuhan incident has exposed three critical gaps in autonomous taxi technology: the lag in evolving from fail-safe to fail-operational design, the danger of emergency systems sharing the same communication path as vehicle control, and the question of operator transparency and accountability when things go wrong.

The purpose of autonomous driving is not just to move — it's to stop safely and call for help when needed. On a Wuhan night, that premise collapsed. As nations integrate this technology into their societies, the lesson from Wuhan is one they cannot afford to ignore.

Sources