Waymo Suspends Robotaxi Service in Five US Cities After Vehicles Drive Into Flooded Roads

On April 20, 2026, an unoccupied Waymo robotaxi in San Antonio encountered a flooded stretch of road. Its sensors registered the water. It slowed down. Then it kept driving — straight into Salado Creek, where floodwaters swept the vehicle downstream. Crews recovered the car four days later along the Greenway Trail system near Pletz County Park [1][2].

That single incident triggered a chain of events that has now forced Waymo, the most commercially advanced autonomous vehicle operator in the world, to suspend service across five cities, recall its entire fleet of 3,791 vehicles, and acknowledge publicly that it has no permanent fix for a problem as old as roads themselves: standing water [3][4].

The Incidents: A Pattern, Not an Anomaly

The Salado Creek disaster was not an isolated failure. Two weeks earlier, on approximately April 4, another unoccupied Waymo vehicle had to be pulled from high water at McCullough Avenue and Contour Drive in San Antonio [2]. That first incident drew little public attention. The second one — a car swept into a creek — made the problem impossible to ignore.

Waymo filed a voluntary recall with the National Highway Traffic Safety Administration on April 30, covering all 3,791 vehicles equipped with its fifth- and sixth-generation automated driving systems across Phoenix, San Francisco, Los Angeles, Austin, San Antonio, and Atlanta [1][3]. The recall filing inadvertently revealed that Waymo's fleet had grown to nearly double the 2,000-vehicle threshold the company had acknowledged in September 2025 [1].

The company deployed an over-the-air software update to the entire fleet and implemented interim restrictions, including limiting vehicle access to areas where flash flooding might occur [3]. San Antonio service, which had been paused since late April, was preparing to resume.

Then, on May 21, storms swept through Atlanta. A Waymo robotaxi drove into a flooded intersection and got stuck for approximately one hour before being recovered [4][5]. The software patch, deployed less than two weeks earlier, had failed.

By May 22, Waymo had suspended service in five cities: Atlanta, Austin, Dallas, Houston, and San Antonio [4][6].

Source: NHTSA, TechCrunch, Electrek

Data as of May 22, 2026CSV

Why Sensors Can't See Water

The technical failure at the heart of this crisis points to a structural limitation in autonomous vehicle perception, not a simple software bug.

Waymo's vehicles carry the most sensor-rich commercial autonomous vehicle stack in regular public operation: a combination of lidar (which maps the environment using laser pulses), radar, and cameras [7]. Yet this array has a fundamental weakness when it comes to water.

Standing water presents two overlapping perception problems. First, its specular (mirror-like) surface reflects lidar pulses unpredictably — sometimes returning a signal that looks like solid ground, sometimes returning nothing at all. Second, camera-based depth estimation degrades sharply in heavy rain, making it difficult to distinguish between a shallow puddle and a road submerged under two feet of water [7][8].

The NHTSA recall filing identified the flaw as a software issue: the system "could cause robotaxis to continue driving and not stop even upon detecting a potentially untraversable flooded lane" [3][4]. In other words, in some cases Waymo's perception stack did detect the water — the vehicle slowed down in the Salado Creek incident — but the decision-making layer failed to classify it as an impassable hazard and route around it.

A further complication: Waymo's system relies in part on National Weather Service alerts to restrict vehicle access during severe weather events. In the San Antonio incidents, flooding was already occurring before any flash flood warning, watch, or advisory had been issued [7]. The system's reliance on external data sources that may lag behind real-world conditions created a gap between what was happening on the road and what the vehicle's operational constraints accounted for.

How Rivals Compare

Waymo's flood failures raise questions about the broader autonomous vehicle industry. No major competitor has publicly reported a comparable incident — but that comes with caveats.

General Motors' Cruise, which suspended its own robotaxi operations in late 2023 after a pedestrian-dragging incident in San Francisco, operated in far fewer weather-variable markets before its shutdown. Aurora Innovation, which focuses on autonomous trucking, has emphasized its Fault Management System — a secondary computer that can take over if the primary system fails — and built-in sensor redundancies [9]. But Aurora's highway-focused operational domain exposes its vehicles to different flood risks than Waymo's urban routes, making direct comparisons difficult.

The underlying physics problem — lidar's unreliable interaction with water surfaces and cameras' degraded depth perception in rain — affects every company using these sensor modalities [10]. Waymo is the first to encounter the problem at commercial scale because it is the only company operating at commercial scale. As of spring 2026, Waymo was completing approximately 500,000 paid robotaxi rides per week and targeting one million weekly rides by year-end [1][5].

Source: Electrek, Waymo filings

Data as of May 22, 2026CSV

The Regulatory Gap

No existing federal regulation specifically requires autonomous vehicles to detect and avoid flooded roadways.

NHTSA's framework for Automated Driving Systems covers SAE Levels 3 through 5. For Level 4 vehicles like Waymo's — which operate without a human driver within a defined operational design domain (ODD) — the framework specifies that the ODD may include prescribed weather conditions, road conditions, and preselected routes [11]. But the standards are voluntary, not binding, and they do not enumerate specific hazard types like flooding that must be detected.

The recall process itself illustrates the gap. Waymo's filing described the flood issue as "an area of improvement regarding untraversable flooded lanes specific to higher-speed roadways" [1][3]. NHTSA accepted the voluntary recall but flagged that Waymo had not yet provided a permanent fix. The agency has authority to demand a remedy but has not publicly set a deadline for Waymo to deliver one.

Several states, including California and Arizona, have their own autonomous vehicle testing and deployment permits, but none include flood-specific requirements. The result is that the most commercially advanced robotaxi fleet in the world operated for months in flood-prone Texas cities without any regulatory mandate to handle a hazard that kills an average of 103 Americans per year [12].

The Human Driver Comparison

Critics of the Waymo suspension have argued that human drivers routinely make the same mistake — and die doing it.

The National Weather Service's "Turn Around, Don't Drown" campaign exists precisely because human judgment fails catastrophically around floodwater. Over half of all flood-related drownings in the United States occur when a vehicle is driven into hazardous flood water [12]. In 2024, driving contributed to 36% of 181 flooding deaths; in 2023, it accounted for 54% of 78 flooding deaths [12][13]. Just 12 inches of rushing water can carry away most cars, and two feet can sweep away SUVs and trucks [12].

Waymo's overall safety record remains strong by comparison. Across 170.7 million rider-only miles driven without a human behind the wheel, Waymo has recorded 0.02 serious-injury-or-worse incidents per million miles, compared to 0.22 for human drivers — a 92% reduction [14][15]. The company also reports 83% fewer crashes involving airbag deployment and 82% fewer crashes causing any injury [14].

Source: Waymo Safety Impact Report

Data as of May 1, 2026CSV

But advocates for stricter AV oversight counter that the comparison is misleading in this context. Human drivers who enter floodwater are making an individual, real-time risk assessment — often a bad one — based on visual cues. An autonomous system that cannot reliably distinguish between a puddle and a creek represents a systemic failure affecting thousands of vehicles simultaneously. When a human driver makes a mistake in a flood, one car is at risk. When Waymo's software has a flaw, 3,791 vehicles share that flaw [3][4].

Who Loses When the Cars Stop

The five-city suspension removed robotaxi service from markets where some riders had come to depend on it. Waymo's service areas in Texas cities like Austin, Dallas, Houston, and San Antonio include neighborhoods with limited public transit options.

Research has shown that approximately 4.5 million people across 52 U.S. cities live in "transit deserts" — areas where residents, who are disproportionately low-income people of color, have fewer transportation options than they need [16]. Waymo has acknowledged this dynamic; in February 2025, the company announced a public transit credit program in Los Angeles aimed at supporting mobility equity [16].

When robotaxi service disappears overnight due to a weather event, riders in well-connected urban cores can switch to conventional ride-hailing, personal vehicles, or public transit. Riders in areas with sparse alternatives face a harder calculation. The suspension's impact is uneven by design: Waymo's service areas were chosen for commercial viability, not equity, and the communities most reliant on the service may be the ones least equipped to absorb its sudden absence.

Specific data on how many daily rides were cancelled during the suspension has not been publicly released. Waymo was completing roughly 500,000 paid rides per week across all markets before the disruptions [1]. The five affected cities represent a significant share of the company's geographic footprint, particularly in Texas, where four of the five suspended markets are located.

Liability and Legal Exposure

No passengers were aboard any of the vehicles involved in the flood incidents — a fact that limits Waymo's immediate legal exposure but does not eliminate it.

Under product liability law, when a company places an autonomous vehicle on public roads, it accepts responsibility for what happens while the vehicle is in autonomous mode [17]. If a passenger had been injured in the Salado Creek incident, Waymo would face potential claims under theories of both negligence and strict liability, with the software's failure to stop before entering floodwater serving as evidence of a defective product [17][18].

Liability in AV incidents often involves multiple parties. The AV operator (Waymo) bears primary responsibility for the vehicle's routing decisions. But claims could also extend to the city or county responsible for road drainage if inadequate infrastructure contributed to the flooding, and to component suppliers if a specific sensor failed to perform as specified [17][18]. The legal framework for apportioning fault among these parties remains largely untested in court, particularly for flood-specific scenarios.

Insurance models are shifting in response. Coverage is moving from primarily driver-based liability to a hybrid model that includes product liability for autonomous systems and manufacturer responsibility [18]. Waymo's $16 billion funding round, which valued the company at $126 billion, provides a financial cushion — but also makes the company a larger target for litigation if a future flood incident causes injury [1].

What Comes Next

Waymo has not announced a timeline for resuming service in the five suspended cities. The company has stated it is "working to implement additional software safeguards," including refining extreme weather operations and further restricting access to flash-flooding-prone areas [3][6].

NHTSA has noted that Waymo has not provided a permanent fix for the flood detection issue. The interim software patch — which was supposed to prevent vehicles from entering detected floodwater — failed in Atlanta on May 21, just nine days after it was deployed fleet-wide [4]. Whether NHTSA will impose a formal deadline or additional requirements remains unclear.

The broader question is whether the flood problem is solvable with software alone. If the underlying issue is that lidar and cameras cannot reliably characterize water surfaces — a physics limitation, not a code limitation — then Waymo may need to add new sensor modalities, integrate real-time hydrological data feeds, or accept permanent geographic and seasonal restrictions on its service areas [7][8].

For now, the company that has driven more autonomous miles than any competitor, that has reduced serious crash rates by 92% compared to human drivers, and that was on track to deliver one million rides per week by year's end is grounded in five cities by a hazard that every human driver in Texas learns to respect: when the road fills with water, stop.