A Classroom for Robots: When Waymo's Driver Failed a Basic Test

Waymo promotes a core principle of its autonomous technology: the entire fleet learns from the experience of a single vehicle. Yet in Austin, Texas, that learning process hit a persistent snag. Over several months, Waymo's robotaxis repeatedly failed to stop for school buses with activated red lights and extended stop arms, a legal requirement. School district officials documented at least 19 such illegal passes.

The company issued a federal software recall in December 2024, acknowledging the issue to safety regulators. Despite this fix, the incidents continued. In an attempt to help, Austin Independent School District officials dedicated a half-day in mid-December to a special data-gathering session. They assembled multiple bus models in a parking lot so Waymo could collect detailed information on their light systems.

The effort did not yield immediate results. By mid-January, the district reported several more violations. An official noted that while human drivers typically learn from a single citation, the automated system appeared not to, despite software updates.

Experts point to a known difficulty for self-driving systems: recognizing specific, context-dependent signals like flashing bus lights and long, thin stop arms. 'If [the company] didn't fix this a few years ago, the more they drive, the more it’s going to be a problem,' said Missy Cummings, a former NHTSA safety advisor and autonomous vehicle researcher at George Mason University.

The National Transportation Safety Board is investigating. A preliminary report on one January incident found a remote human assistant incorrectly told the vehicle the bus's signals were off. This episode underscores a fundamental hurdle: programming software to navigate the subtle, exception-filled rules of human roadways, where a stop sign on a bus carries different implications than one on a street corner.

Source: Wired gear