Why didn’t Waymo learn to stop buses?

Waymo’s school-bus training didn’t translate into safer behavior

A school district tried to help train Waymos to stop for school buses, expecting the usual advantage self-driving advocates cite: learning from mistakes across vehicles. The effort didn’t work.

The underlying idea was straightforward—improve bus behavior by providing more targeted examples and training signals. But the attempt failed to produce the desired outcome, leaving the district with an example of how “collective learning” can break down in real-world edge cases.

The story frames the issue around how Waymo describes its system: the Waymo Driver “learns” from driving experience. The district’s experience suggests that, even with that promise, some safety-critical interactions—like the specific procedures around school buses—may not be captured adequately by the training approach used.

This matters because school transportation is one of the most sensitive contexts for autonomous driving: predictable compliance with stopping rules affects not only the vehicle’s trajectory but also the safety expectations of children, parents, and bus operators.

The episode highlights a broader challenge for autonomy deployments:

• Not every failure mode is easily teachable just by adding training data.
• Policy and interpretation details (what counts as the right cue, at the right time) can be as important as raw experience.

Until those gaps are closed, districts and regulators may continue to face uncertainty when deciding whether to trust autonomous systems in high-stakes environments.