Waymo issued a voluntary NHTSA software recall for ~3,800 robotaxis using 5th/6th gen ADS after vehicles drove onto flooded roads and stalled.
Key Takeaways
Trigger incident: April 20, a Waymo AV in San Antonio entered a flooded road and was swept into a creek; no passengers aboard, but it opened an NHTSA probe.
Recall covers all U.S. Waymo vehicles on 5th and 6th generation automated driving systems; fix is an OTA software update, not a physical service action.
Waymo has applied interim mitigations: operational geofencing during extreme weather to avoid flash-flood-prone areas.
San Antonio service remains suspended; Waymo says it is readying to resume public rides.
Pattern of edge-case failures: prior incidents include failure to yield to school buses in Austin and gridlock during SF power outages in December.
Hacker News Comment Review
Commenters flagged that distinguishing reflective wet pavement from traversable standing water is genuinely hard for both humans and AV sensor stacks; a dedicated water-depth sensor was suggested as a direct fix, with the tradeoff of over-caution at shallow puddles.
The Texas geography context matters: flat terrain, limestone substrate, and dry creek “low water crossings” make flash flooding unpredictable and common, raising the difficulty bar beyond typical AV test environments.
Several commenters pushed back on the word “recall,” noting it is a regulatory term that here means an OTA patch while vehicles are pulled from service, not a physical return; some argued NHTSA should modernize the terminology for software-updatable vehicles.
Notable Comments
@Animats: Notes DARPA Grand Challenge teams solved this in 2005 with onboard water sensors; proposes cautious entry to probe depth before committing.
@Zigurd: Suggests Google’s map scale enables inference-based flood detection combining vehicle deceleration, steering correction, and computer vision without dedicated hardware.
