Waymo Recalls 3,791 Autonomous Vehicles: How a Robotaxi Drove Into Flooded Roads

Waymo Recall: A Robotaxi Drives Into Flooded Roads

Alphabet's autonomous driving subsidiary Waymo recently announced a recall of its self-driving software after the system failed to prevent a vehicle from entering a flood-submerged road. The recall affects 3,791 vehicles equipped with Waymo's fifth- and sixth-generation systems, once again putting the decision-making shortcomings of autonomous driving systems in extreme weather conditions under the spotlight.

According to the defect report Waymo filed with the National Highway Traffic Safety Administration (NHTSA), an unoccupied Waymo robotaxi "encountered an impassable flooded road segment" during operation. Rather than stopping or rerouting in time, the vehicle continued driving into the waterlogged area. This incident directly triggered the software recall.

Recall Covers 3,791 Vehicles; OTA Remote Fix Is a Highlight

Affected Vehicles and System Versions

This recall covers Waymo's currently operational fifth- and sixth-generation autonomous driving systems, totaling 3,791 vehicles. Waymo's autonomous driving system has undergone multiple iterations. The fifth-generation system (Driver 5th Gen) is mounted on the Jaguar I-PACE electric SUV, employing a fusion approach of 29 cameras, LiDAR, and radar sensors with a perception range exceeding 300 meters. The sixth-generation system is a new platform co-developed with Geely, built on the Zeekr vehicle platform, featuring significantly reduced sensor costs while further enhancing perception capabilities—seen as a critical step for Waymo to achieve large-scale commercial operations. Together, these two generations form the core technology platform for Waymo One robotaxi services in San Francisco, Phoenix, Los Angeles, and other cities. The recall scope covers virtually the entire operational fleet.

Software Recall, Not Hardware Recall

The core issue in this recall lies in the autonomous driving software's environmental perception and decision-making modules—the system failed to correctly identify flood conditions and respond safely. Under normal logic, when an autonomous driving system detects a road submerged in water ahead, it should automatically reroute or pull over to stop. However, this software exhibited a judgment blind spot when facing flood scenarios.

The good news is that this is a software recall, and Waymo can complete the fix through OTA (Over-The-Air) remote updates without vehicles needing to return to service centers. OTA remote update technology was first widely adopted for smartphone system upgrades before Tesla introduced it to the automotive industry, where it gradually became an industry standard. Its core principle involves pushing software update packages to vehicles via cellular networks or Wi-Fi, with the onboard computing platform handling download, verification, and installation. For autonomous vehicles, OTA can not only fix software defects but also continuously optimize perception algorithms, update HD map data, and improve decision-making logic. Unlike traditional automotive recalls that require owners to bring vehicles to dealerships for repairs, OTA recalls can cover an entire fleet within hours, dramatically reducing both time and economic costs. However, OTA also introduces new cybersecurity challenges—how to prevent malicious tampering during the update process remains an ongoing industry concern. This approach is both efficient and cost-effective, serving as a textbook example of the recall model in the software-defined vehicle era.

Extreme Weather Remains a Technical Achilles' Heel for Autonomous Driving

Floods, Heavy Rain, Heavy Snow—A Nightmare for Sensors

Extreme weather conditions such as floods, heavy rain, and heavy snow have long been among the most challenging obstacles for autonomous driving technology. Specifically, these scenarios present three layers of difficulty for autonomous driving systems:

• Severely degraded sensor performance: Rain splashes and standing water on road surfaces can seriously interfere with the perception accuracy of LiDAR and cameras, causing deviations in obstacle detection and road surface recognition. LiDAR constructs a 3D point cloud map of the surrounding environment by emitting laser pulses and measuring reflection time, making it one of the most critical sensors in current L4 autonomous driving systems. However, LiDAR has significant limitations in extreme weather: raindrops in heavy rain scatter laser signals, producing massive "noise points" that cause the system to misidentify rain curtains as obstacles; waterlogged road surfaces can reflect lasers like mirrors, causing ground detection failures; heavy fog and dust can severely reduce effective detection range. To address these challenges, companies like Waymo employ multi-sensor fusion strategies—complementarily fusing data from LiDAR, millimeter-wave radar (unaffected by rain and fog but with lower resolution), and cameras (providing semantic information but affected by lighting conditions) to improve perception robustness in adverse weather. Yet this flood incident demonstrates that even multi-sensor fusion approaches still have perception blind spots in certain extreme scenarios.

• Difficulty assessing road conditions: Floods can completely alter road passability, and water depth cannot be accurately estimated solely through visual sensors, posing an enormous challenge for decision-making systems.

• Severely insufficient training data: Extreme weather represents an extremely small proportion of autonomous driving training datasets, constituting a typical "long-tail scenario" where model generalization capability in these edge cases is inherently weak. In machine learning, "long-tail distribution" refers to the collection of numerous low-frequency events—each individual event has an extremely low probability of occurrence, but the variety is vast and the total volume is considerable. For autonomous driving, everyday sunny-day urban driving belongs to "head scenarios" with abundant training data and excellent model performance; while floods, hail, and wildlife crossing highways belong to "long-tail scenarios" that may appear only a handful of times across billions of miles of training data. This results in insufficient learning samples for the model in these scenarios, leading to inadequate generalization capability. The industry currently addresses long-tail data gaps primarily through simulation (e.g., Waymo's Simulation City platform can generate billions of virtual driving scenarios), adversarial training, and domain adaptation techniques. However, the "sim-to-real gap" between simulation and the real world remains a technical challenge that has not been fully resolved.

Safety Redundancy Mechanisms Urgently Need Strengthening

In this incident, the vehicle happened to be unoccupied, resulting in no injuries. But consider the alternative: if passengers had been inside, the consequences of driving into a flooded area could have been quite severe—a stalled vehicle trapped in water, passengers facing drowning risk—none of these are low-probability events.

This serves as a wake-up call for the entire autonomous driving industry: in extreme scenarios, systems should establish stronger safety redundancy mechanisms, preferring false positives that result in longer detours over risking passage through unknown road conditions. Safety Redundancy is a classic design philosophy from the aerospace industry, with the core concept of ensuring that single-point failures do not lead to catastrophic consequences through multiple independent layers of safety safeguards. In autonomous driving, safety redundancy is typically reflected at three levels: perception redundancy (cross-validation across multiple sensor types), computational redundancy (primary and backup dual computing platforms, with the backup system taking over when the primary fails), and decision redundancy (automatically triggering a Minimal Risk Condition, such as pulling over, when system confidence falls below a threshold). Waymo's system design includes a "Minimal Risk Condition" (MRC) mechanism—when the system determines it cannot safely continue driving, it automatically guides the vehicle to a safe location to stop. However, the problem exposed by this flood incident is precisely that the system failed to identify the flooded road segment as a dangerous scenario requiring MRC activation, indicating that the trigger conditions for safety redundancy mechanisms still need broader coverage.

Behind the Proactive Recall: Industry Regulation and Public Trust

Waymo's Proactive Reporting Sends a Positive Signal

Waymo chose to proactively report the defect to NHTSA and initiate the recall, rather than waiting for regulators to intervene before responding passively. During a critical window period where the autonomous driving industry is still building public trust, this transparent and responsible approach benefits the long-term development of the entire industry. By contrast, concealing problems or delaying responses only exacerbates public distrust of driverless technology.

NHTSA Continues Tightening Autonomous Driving Oversight

NHTSA has noticeably increased its regulatory oversight of autonomous vehicles in recent years, not only requiring companies to proactively report accidents and defects involving autonomous driving systems but also launching investigations into multiple companies. The National Highway Traffic Safety Administration (NHTSA) is the core federal agency responsible for automotive safety regulation in the United States. In June 2021, NHTSA issued a Standing General Order (SGO) requiring all vehicle manufacturers and operators with L2 or higher autonomous driving or advanced driver assistance systems to report accidents involving serious injuries or fatalities within 24 hours, and other relevant accidents within 30 days. To date, NHTSA has launched formal investigations into Tesla Autopilot, GM Cruise, Waymo, and other companies. In 2024, Cruise was fined and had its operating permit suspended by NHTSA after its robotaxi dragged a pedestrian in San Francisco, becoming a landmark regulatory case for the industry. Waymo's proactive defect reporting stands in stark contrast to Cruise's alleged concealment of accident details at the time, also reflecting the industry's gradual move toward greater transparency under regulatory pressure. This Waymo recall event will serve as another important reference case for regulators evaluating autonomous driving technology maturity and may drive the development of more detailed extreme weather testing standards.

Conclusion: How Far Is Autonomous Driving From "All-Weather Operations"?

This Waymo software recall demonstrates a reality: autonomous driving technology still has a clear gap before achieving truly "all-weather, all-scenario" commercial operations. In routine scenarios such as daily commuting and urban roads, Waymo's robotaxis have accumulated millions of miles of safe driving records, performing even better than human drivers in some metrics. But when encountering long-tail scenarios like floods, ice and snow, or sandstorms, the system's robustness reveals its shortcomings.

For Waymo, the ability to quickly fix issues via OTA is itself a demonstration of technical capability. But the more critical challenge lies in: how to fundamentally improve autonomous driving systems' perception and response capabilities in extreme scenarios? This not only determines whether Waymo can continue expanding its operational footprint but will also decide when the entire autonomous driving industry can truly earn widespread public trust.

Key Takeaways

• Waymo recalled software across 3,791 autonomous vehicles equipped with fifth- and sixth-generation systems
• The recall was triggered by a driverless vehicle inadvertently entering a flood-submerged road segment, with the system failing to correctly identify dangerous road conditions
• This is a software recall that can be fixed via OTA remote updates without vehicles returning to service centers
• Environmental perception and decision-making in extreme weather conditions remain a major challenge for autonomous driving technology
• Waymo's proactive reporting to NHTSA demonstrates industry transparency and helps build public trust