Another one on “Edge Cases”. Waymo, this time

Gift link:

https://www.nytimes.com/2026/08/10/technology/waymo-expansion-edge-cases.html?unlocked_article_code=1.4VA.a_Hs.zR_xQPAGapti&smid=nytcore-ios-share

Waymo Is Growing Faster Than Ever. So Are Its Glitches.

As Waymo deploys more driverless cars to 15 U.S. cities and counting, its vehicles keep encountering new and unexpected situations that they have no script to handle.

What Ms. Peterson experienced is what the driverless car industry calls an “edge case,” which are the unscripted situations that no one trained the robo-taxis to handle. The problem is that edge cases appear to be piling up as Waymo, the leading autonomous car service, rapidly expands.

Owned by Google’s parent Alphabet, Waymo has more than quintupled the number of autonomous cars it has on the road to nearly 4,000 today, up from about 700 early last year. Its vehicles are now in 15 metro areas, having gone fully driverless in Denver, Las Vegas, San Diego and Tampa in the past month alone. Each week, passengers pay for about half a million trips.

And here’s partly why:

More cars in more cities means more chances to meet something the technology has never seen. That’s hard to fix because the cars often do not recognize they are in trouble and so do not request help, said Philip Koopman, a Carnegie Mellon University emeritus professor who specializes in autonomous vehicles.

“This technology struggles to know when it’s in a novel situation,” Dr. Koopman said. “And when it’s in a novel situation, it’s prone to overconfidence and spectacularly stupid failure.” Waymo’s answer has not been to solve the underlying problems, he said, but to hand control of the car to a remote human operator.

4 Likes

This is my recent favorite, where empty Waymo cars swarm a residential cul-de-sac in Atlanta.

DB2

5 Likes

We were driving around the UT Austin campus one day during evening rush hour. A Waymo in front of us was attempting a legal left-hand turn that it was never going to be able to make at that time of day and it did not appear to realize that. Any human would have known to turn left about 100 feet earlier, or 200 feet later, not because of legality but because of ease and chances of success. This is why I don’t understand the push for driverless taxis and city autonomous driving. Highway cruising for road trips, sure. City driving nope. I mean seriously, how much more “data” and “training time” do we need for this, at what cost, and is it really helping us?

5 Likes

Kinda makes one think unsupervised autonomy for personal vehicles is kinda far away.

The nationwide fleet is going to sleep on this march of 9s challenge for a few more years, at least.

Heard in 2019

Next year for sure, we will have over a million robotaxis on the road. The fleet wakes up with an over-the-air update. That’s all it takes.”

Heard in 2026

Ideally, you want 99.999999% reliable. It’s just the march of nines of reliability is, I think, the only thing really constraining our growth in Robotaxi.

3 Likes