Waymo's Flood Fix Fails: Software Patch Breaks Down, Service Halted in Six Cities

The most commercially advanced robotaxi fleet in the world cannot reliably operate in the rain. That uncomfortable sentence is now the working conclusion from Waymo’s most turbulent month since launching its public service — a story that began with a single car swept into a San Antonio creek in April and escalated, by the third week of May, into a multi-city service shutdown with no permanent remedy available.

How a Creek in Texas Became a National Safety Crisis

On the evening of April 20, an unoccupied Waymo vehicle operating in San Antonio encountered a flooded section of road during a heavy downpour. Rather than stopping, the car slowed briefly and then drove forward into the standing water. The floodwaters swept the vehicle into Salado Creek. No passengers were aboard, but the footage circulated widely — an embarrassing visual for a company that has staked its commercial reputation on the premise that its autonomous system makes better decisions than human drivers.

Waymo filed a voluntary software recall with the National Highway Traffic Safety Administration on April 30. NHTSA sent a formal inquiry letter to the company on May 11, and Waymo publicly confirmed on May 12 that it was recalling all 3,791 vehicles equipped with its 5th- and 6th-generation Automated Driving Systems — essentially its entire commercial fleet. The fix was delivered over-the-air: no vehicles needed to visit a service center. Waymo described the update as increasing “weather-related constraints” and improving how the vehicles interpreted flooded road data pulled from updated maps.

The recall covered roughly three weeks of exposure. According to NHTSA filings, the software had a flaw that allowed vehicles to reduce speed but still proceed onto standing water at higher-speed roadways — a behavior that would be considered unacceptably dangerous if executed by a human driver.

The Patch That Wasn’t

Less than two weeks after the OTA fix was pushed to every vehicle in the fleet, Waymo was back in the headlines for the same reason.

On May 21, as severe thunderstorms dropped two to three and a half inches of rain across Atlanta in a matter of hours, two separate Waymo vehicles drove into flooded streets in the city. One of the cars — an unoccupied unit operating autonomously in Midtown Atlanta — became stuck in standing water and required human recovery after approximately one hour. Video of the stranded vehicle spread rapidly on social media.

Waymo acknowledged the failure publicly: the patch it had deployed in response to the San Antonio incident had not prevented the Atlanta occurrence. The company suspended service in Atlanta, Dallas, Houston, San Antonio (which had already been paused since late April), and Austin. Nashville was subsequently added to the list, bringing the total to six cities with halted operations.

The company disclosed in a statement that it had not yet developed a “final remedy” for the underlying software problem. That phrasing — borrowed from the regulatory vocabulary of vehicle recalls — carries significant weight. It means the most commercially successful autonomous vehicle operator in the United States is operating its urban fleet without a complete solution to a known safety defect.

Two Problems, One Fleet

The flooding incidents were not the only safety issue to surface on May 21. On the same day, Waymo announced a separate, unrelated suspension: it was temporarily halting all freeway rides across its operating markets — including San Francisco, Los Angeles, Phoenix, and Miami — while engineers worked to improve the system’s behavior in highway construction zones.

The two announcements arriving simultaneously gave the impression of a fleet under pressure on multiple fronts. Waymo has spent the past three years projecting confidence about its expanding geographic footprint: San Francisco, Phoenix, Los Angeles, Miami, Nashville, Atlanta, and several Texas cities had all been brought into service. The dual suspension was a jarring reversal of that narrative.

The freeway situation reflects a different class of challenge. Construction zones present adversarial environments for autonomous systems: irregular lane markings, temporary barriers, human flaggers, and unpredictable routing create conditions that fall outside the clean, well-mapped scenarios on which current ADS generations are trained. Waymo did not specify when freeway rides would resume or what changes to the system were being implemented.

Regulatory and Competitive Fallout

NHTSA has been actively monitoring Waymo’s recall status and has opened broader inquiries into its autonomous driving systems. The National Transportation Safety Board has also flagged earlier collision data involving Waymo vehicles for review. The simultaneous suspension in six cities increases the likelihood that federal scrutiny will intensify.

For competitors, the timing is double-edged. Tesla CEO Elon Musk had been loudly positioning the Cybercab robotaxi as a superior alternative to what he characterized as Waymo’s expensive, geofenced approach. But Tesla’s Austin robotaxi pilot — operating 44 vehicles as of early May, against a target of 500 by year’s end that has not been met — is in no position to absorb displaced Waymo riders. Zoox, Aurora, and other AV companies are watching closely: the operational failures expose a systemic challenge with edge-case weather handling that affects every current-generation ADS, not just Waymo’s.

What This Means for Autonomy at Scale

The flooding crisis is, at its core, a data and systems problem. Autonomous vehicles navigate using high-definition maps combined with real-time sensor data from cameras, radar, and lidar. The challenge with standing water is that it can appear on roads that were dry when the map was last updated, move unpredictably during active storms, and — critically — look different to sensors depending on lighting conditions and the depth of the water.

The April 20 incident revealed that Waymo’s pre-existing sensor fusion logic was not adequate to override a flooded-road entry on high-speed roads. The May 12 OTA patch attempted to address this by tightening weather-related operational constraints and updating map data, but the Atlanta recurrence suggests the fix was either incomplete, insufficiently tested under real storm conditions, or failed to cover a broader class of flooding scenarios.

Waymo has not disclosed the timeline for a permanent fix or what engineering work is required to achieve it. Until that solution exists, the company’s fleet will continue operating under manual suspension protocols during severe weather — a contingency that undermines the “no driver needed” promise that defines the entire robotaxi value proposition.

The question the industry will be watching: how long does a company keep shutting down entire markets every time it rains before the public, regulators, or both, decide that the technology is not yet ready for the density of deployment Waymo has pursued?

The answer, increasingly, looks like it will come from Washington before it comes from Mountain View.