I was sitting at my desk when the flood of videos arrived: stopped cars, flashing hazard lights, passengers stepping onto elevated asphalt with trucks whistling by. I felt the room tilt; this was not a stalled navigation app or a polite traffic jam. It was hundreds of robotaxis frozen in place—traffic cones with human beings inside.
I’ll tell you what I’d do if you ever find yourself in one of these boxes. I’ve ridden with Waymo engineers, argued with product leads at Tesla events, and watched footage from Baidu’s Apollo Go collapse across Wuhan’s ring roads. You need clear steps, and you need context so you stop guessing while the minutes pile up.
Cars stopped on elevated ring roads. What that looked like in real time.
On March 31 in Wuhan, Baidu’s Apollo Go fleet began to stall just before 9:00 p.m. Videos show dozens of robotaxis parked in lane, some on expressways where there is no shoulder and no safe place to step out. Witnesses reported collisions, panic, and a wash of confused messages across social media.
According to Wired and Reuters reporting, about 100 vehicles experienced a “system failure” that left passengers waiting inside. Chinese TV interviews collected by CarNewsChina read like micro-thrillers: Mr. Lu waiting nearly an hour for help while trucks roared by, Ms. Zhou told not to open her door, and, in a moment that reads worse on paper than it must have felt in the car, passengers still charged full fares after being held captive for hours.
如果你在为工作中的失误而懊恼的话
不妨想想这个让全武汉的无人驾驶萝卜快跑罢工还停在路中间的团队（） pic.twitter.com/K0RFBadSkH— 𝓙𝓪𝓬𝓲𝓮 𝓩𝔂𝓽 (@Jaciezyt) March 31, 2026
Baidu’s initial statement calls it a “system failure.” Gizmodo noted police translations saying everyone “safely disembarked.” Yet the footage and passenger interviews suggest the outage was a cork sealing off the ring road: no shoulder, no quick exit, and phone calls that dropped or routed to automated systems.
What should I do if my robotaxi stops?
If you’re in the car, stay calm and obey any on-screen warnings. Try the in-car SOS button and the app’s emergency contact. If those fail, call local emergency services immediately and provide your precise location—elevated ring roads have limited access points. Record timestamps and take photos; companies and investigators will want that record.
Passengers tried customer service. Here’s how it actually unfolded.
A passenger tried the backseat screen. It dropped the call. The in-car SOS button was, in one account, “completely useless.” The official hotline routed some riders to a queue that never answered for nearly an hour.
That delay is why this matters. Robotaxi operators like Baidu and Waymo tout aggregated safety figures—Waymo’s numbers often outperform human drivers, and Tesla claims ambitious robotaxi goals—but aggregated metrics obscure single-point failures. An overloaded human feedback loop can cause a slow-down; a systemic outage can strand dozens in hazardous spots.
NEW: Dozens of robotaxis by Baidu stopped on the road in Wuhan, China, causing crashes on highways and trapping passengers in the cars—some for more than an hour. One passenger told me it took her 30 minutes to even connect to a customer representative. Here’s a dash cam video of one crash.
— Zeyi Yang 杨泽毅 (@zeyiyang.bsky.social) March 31, 2026 at 6:33 PM
Are robotaxis safe?
Short answer: statistically they can be, but statistics hide single events that matter to you. Waymo’s slow, cautious approach produces strong aggregate safety numbers; Tesla’s public robotaxi ambition has raised questions about readiness. Baidu’s outage shows a different failure mode—mass paralysis rather than isolated collisions—and that matters because it changes the risk you face in the car.
People left the cars. Then the bills arrived.
Ms. Zhou and Mr. Lu were allowed off elevated roads after police and company staff arrived—hours after the outage began. They walked away shaken, and in Ms. Zhou’s case, charged the full fare. That billing detail landed in reports like a small but venomous sting.
You can think of money as a second axis of harm: time plus risk plus cost. If companies continue to bill full fares during outages, regulatory pressure and consumer lawsuits will follow—remember the high-profile settlements in other self-driving incidents reported by outlets like the LA Times and The New York Times. Those headlines move markets and product roadmaps.
Who is liable if a robotaxi malfunctions?
Liability sits with a mix of the operator, the stack provider, and local regulators. If Apollo Go owns the fleet and the software, Baidu bears much responsibility; if a third party supplies mapping or sensors, blame spreads. Legal systems are still catching up, and each case will hinge on evidence: logs, sensor records, and service transcripts.
Investigative outlets—Wired, Reuters, Gizmodo—will hunt for logs. Regulators will ask for post-incident reports. You should assume the company will preserve telemetry, so your photos and audio are powerful complements to official data.
So what changes after a mass outage? Three practical shifts I expect.
Operators will face operational fixes, policy scrutiny, and user mistrust. Practically: stronger in-car fail-safes, better human triage lines, and clearer refund policies. That’s the minimum customers will demand after watching people sit on highways for two hours—about a feature film’s runtime—while customer service falters.
Companies like Baidu will issue technical postmortems; platforms such as Apollo Go will tweak escalation flows. Waymo and Tesla will study the incident too—competitors do not enjoy each other’s pain, but they learn from it. Regulators will press for standards around triage, access, and mandatory egress procedures on elevated roads.
If you ride robotaxis, carry a charger, screenshot ride details, and know your local emergency number. Record the time your ride stopped and any failure messages on the screen. Those simple acts are the difference between a frustrating story and a usable legal record.
The outage lasted roughly two hours—long enough to be its own contained thriller and short enough that the narrative moves fast online. We will see post-incident PR, technical analyses, and maybe a Blumhouse-style dramatization, but the harder questions are already out of the car: will operators update their human-response systems, and will regulators force meaningful standards?
When a fleet turns immobile, who pays the human cost and how will you prove it?