Waymo Halts Service in Five Cities: Flood Patch Fails Weeks After Recall, Permanent Fix Missing

Waymo suspended robotaxi service across five cities on May 21, 2026, after a software patch it pushed to its entire 3,791-vehicle fleet less than two weeks earlier failed to prevent another autonomous vehicle from driving into standing water — this time in Atlanta, where the car stalled in a flooded Midtown street for approximately an hour before it could be recovered. The company acknowledged it still has no permanent fix for the problem, meaning the most commercially advanced robotaxi fleet in the world cannot reliably operate during heavy rain, and riders in five markets have no recourse when storms arrive.

Waymo halted service in Atlanta, Dallas, and Houston as severe weather swept through the region on May 21, with San Antonio already suspended since late April following a separate incident in which an unoccupied vehicle was washed into a creek; Austin was also paused as a precaution. That came on the same day that Waymo separately suspended freeway rides in San Francisco, Los Angeles, Phoenix, and Miami after its vehicles were observed struggling in highway construction zones. The National Highway Traffic Safety Administration told TechCrunch it "is aware of this incident, is in communication with Waymo, and will take appropriate action if necessary."

Patch Deployed Nationwide in May Did Not Prevent Atlanta Failure

The Atlanta incident is the direct sequel to what was supposed to be a solution. On April 20, a Waymo robotaxi in San Antonio encountered a flooded section of road on a 40 mph corridor, slowed but did not stop, and was swept into Salado Creek. The car was unoccupied and recovered days later. No one was injured, but the event triggered a voluntary recall filed with the NHTSA covering 3,791 vehicles equipped with Waymo's fifth- and sixth-generation autonomous driving systems — the company's entire commercial fleet.

Waymo shipped an over-the-air software update that placed, in the language of NHTSA recall documents, "restrictions at times and in locations where there is an elevated risk of encountering a flooded, higher-speed roadway." The company conceded in those same documents that it was still developing a "final remedy" — meaning the patch was a stopgap, not a fix. The Atlanta incident confirmed the stopgap's limits: flooding was already occurring before the National Weather Service had issued any flash flood warning, watch, or advisory, and Waymo's system relies in part on those formal alerts to restrict vehicle access during severe weather. The same interval — about a week and a half between the recall patch and the Atlanta failure — shows how quickly the interim measure ran out of runway.

Sensor Architecture Cannot Estimate Flood Depth, Researchers Say

Waymo's vehicles use a combination of LiDAR, radar, and cameras — the most sensor-rich commercial autonomous vehicle stack in regular public operation. That sensor array is precisely the reason the flood failure carries weight beyond Waymo: it points to a structural limitation, not a configuration error.

"Neither are particularly well-suited to handle, to estimate how deep a particular puddle is. That can be really a problem for flooding," said Glen Chou, a Georgia Tech professor who studies autonomous vehicle perception, referring specifically to LiDAR and radar. Chou said companies such as Waymo learn from unusual "corner cases" using real-world sensor data, and that the path forward involves either programming explicit fallback behaviors when those corner cases arise or building training datasets that prescribe correct behavior. "Knowing more about the kind of corner cases that can occur in driving, being able either to program some kind of fallback when those corner cases arise or having data that tells you what you should do," Chou said.

Srinivas Peeta, who leads Georgia Tech's Autonomous and Connected Transportation Lab, said the gap between public expectation and autonomous vehicle capability is not always visible until something goes wrong. "It is puzzling for humans to say, 'Oh, these are autonomous vehicles. They are supposed to be more advanced. They should be much more intelligent than us,'" Peeta said. "But that's where the difference occurs."

Water depth estimation sits at the intersection of two known hard problems in autonomous vehicle perception: the specular surface of standing water reflects LiDAR pulses unreliably, and camera-based depth estimation degrades sharply in heavy rain. The San Antonio incident made clear the problem is not just recognition — the system detected the floodwater — but decision architecture: the vehicle's software registered the hazard and reduced speed rather than halting, because on a 40 mph road there was no hard-stop condition for water in its path.

Third Recall in Two Years Joins Two Active Federal Investigations

The flood recall is Waymo's third since February 2024. The first covered 444 vehicles after two robotaxis in Phoenix separately struck the same improperly towed vehicle. The second, filed in May 2025, covered 1,212 vehicles involved in low-speed collisions with stationary barriers including parking gates and telephone poles.

The current recall arrives alongside two active federal investigations into separate failure modes. The NHTSA is investigating a January 23, 2026, incident in which a Waymo robotaxi struck a child near Grant Elementary School in Santa Monica, California, during morning drop-off hours. The vehicle braked from approximately 17 mph to under 6 mph before contact, and the child sustained minor injuries. The National Transportation Safety Board opened a parallel investigation.

The same two agencies are also investigating Waymo's school bus failures. The Austin Independent School District documented at least 24 instances of Waymo vehicles illegally passing stopped school buses during the 2025-2026 school year, including incidents that occurred after Waymo deployed a prior software update to address the problem. On May 15, the NHTSA sent Waymo a second document request in the school bus investigation after concluding that the company's initial response required further data.

Waymo Targets 1 Million Weekly Rides by Year-End as Failures Pile Up

The timing is particularly awkward for Waymo. The company raised $16 billion in February 2026 at a $126 billion valuation and has set a goal of offering 1 million paid rides per week by the end of 2026 — roughly four times its current rate of approximately 250,000 rides per week. The company is simultaneously testing its new Zeekr-built sixth-generation vehicle, called Ojai, and has begun safety-driver testing in London with a wider public launch expected later in 2026.

That expansion context sharpens the problem in two ways. The fleet racing toward a million rides per week still carries an unfixed flood vulnerability during thunderstorm season. And it raises a structural question for the cities being added to Waymo's service map: what weather-condition testing standards apply before commercial deployment in a new market? Neither the NHTSA nor any other federal regulator currently requires an autonomous vehicle operator to demonstrate flood-navigation performance before launching commercial service.

Waymo said it is "working to implement additional software safeguards" and has put interim mitigations in place. The company did not provide a timeline for when a permanent fix would be available or when service would resume in the suspended cities.

For riders in Atlanta, Austin, Dallas, Houston, and San Antonio, the practical answer to "can I get a Waymo in a thunderstorm?" is still no — not because the service opted out, but because its vehicles drove into the water and proved it cannot.