Phil Koopman 11/18/26 newsletter – Waymo Growing Pains and Failed Fixes

Phil Koopman is a leading expert on robotaxi safety issues. Koopman’s 11/18/26 newsletter is incredibly useful because it explains and gives examples of how Waymos are not safely handling rare situations, the problem is not being fixed, and it appears to not be fixable. If you want to know more about robotaxi safety I highly recommend that you subscribe to his newsletter at https://philkoopman.substack.com/subscribe


A big-picture perspective on the persistent problems that Waymo can’t seem to fix — even while they push to scale up deployments

Recently we’ve seen a spate of incidents involving Waymo robotaxis, including a recent California DMV action to kick Waymo vehicles off freeways. But I’m not here to hash out specific incidents. Rather, I want to make two related points:

  1. These incidents aren’t one-offs. They are just the latest pieces in long-term patterns of problems.
  2. Promises that robotaxis will inexorably improve in response to incidents aren’t working out.

A common theme in the robotaxi safety journey is that every incident is an opportunity to learn and improve. Sure, the opportunity is there. But do they really improve? Here are the receipts showing that some patterns of problems are resisting fixes.

For those who want the high level discussion, here is a 10-minute interview I did with KTLA-TV:

The details are below, but I’ll summarize the trends first. In numerous areas, we’re seeing a pattern. Safety-relevant incidents involving Waymo vehicles keep happening despite attempted fixes. We’ve seen a pattern of recalls that did not prevent further incidents for:

  • Improperly passing stopped school buses
  • Driving through flood waters
  • Driving in closed areas of highway construction zones

There are also persistent patterns of incidents in other areas that might yet lead to recalls. Waymo has not attempted fixes that rise to the level of a federal recall, but there is every reason to believe they have tried and failed to fix these problems.

  • Interfering with traffic flow in general, and with emergency responder operations in particular
  • Dangerous behavior around rail tracks: getting stuck at rail crossings and driving on dedicated rail right-of-ways
  • Intruding into road closures on surface roads
  • Problems interpreting manual traffic direction hand signals
  • Driving into dangerous situations that are better avoided

If we’re seeing these patterns with a few thousand robotaxis, what does it look like when we see scale-ups of 10s or 100s of thousands of vehicles? With 100x more robotaxis on the road, once-a-year incidents that nobody noticed become twice per week social media clips.

Infrequent loose ends that might be tolerable with perhaps a dozen vehicles need to get fixed before we see them as constant problems on thousands of vehicles. Waymo’s track record shows that fixes for some types of problems aren’t working. Yet scaling-up proceeds apace.

Below is a list to document just some of the publicly-known incidents in each category. No doubt there are more that Waymo knows about (or ought to know about). And I’m sure I’m forgetting some that should be on the list as well. But I think the lists are long enough to make my point here. There are clear patterns of problems that aren’t being fixed, in many cases despite obvious attempts by Waymo to fix them over extended lengths of time.

What I would like to see is more transparent public reporting of misbehaviors that are seen to form a pattern, especially if those behaviors are supposed to be fixed as part of a recall process. If Waymo says they have fixed passing school buses, they should create a public dashboard of how often it is happening. Nothing in this world is perfect, but a credible dashboard showing consistent improvement over time would go a long way toward fostering public acceptance. We shouldn’t have to scour social media to find out that a promised fix didn’t really work.

Having a transparent public dashboard might also make it easier for regulators to accept that a fix will work and permit cars back on the road with less evidence than would be prudent to require otherwise. Right now, California DMV has said that Waymo needs to convince DMV that freeway construction zone intrusions are fixed before Waymo is allowed back on freeways. But the last recall “fix” didn’t work. So how is Waymo supposed to prove that this time is different? This time for sure! Sounds like a tough sell.

All these issues are a problem only sometimes, and mostly involve a small fraction of robotaxis. However, they happen often enough to be an issue, and at times happen to multiple vehicles making the same consistently wrong behavioral mistake in the same situation. And yes, human drivers make all these mistakes too. But we were promised robotaxis would not make stupid human mistakes. Turns out to be harder than claimed.

The issue here is that we need to stop pretending that robotaxis are immune to repeated mistakes because of a false narrative that every mistake results in a permanent fix. We also need to find a way to be able to trust that all autonomous vehicle companies will actually fix things they say are fixed.

Before I dive into the receipts, there are some of my pieces on related topics that support these ideas:

The following sections list patterns of concerning incidents. In each case it is clear that Waymo knew (or should have known) about these problems. In many cases Waymo acted to mitigate the problems — but the problems kept happening anyway.

Driving into flood waters:

Driving into flood waters risks the car being swept away and/or a passenger being trapped sitting in a car while the water rises.

It seems the fix for now is based on dynamic geofencing around flooded streets and some sort of real-time detection of standing water that provokes a shutdown. That is not the same as teaching the AI driver how to handle standing water appropriately based on conditions.

Passing school buses:

A common traffic law is that cars cannot drive passed stopped school buses that are loading/unloading students on undivided roads. Many school buses, including the ones in reported incidents, have not only flashing lights, but also swing-out stop signs and barrier bars that are deployed when performing loading/unloading.

It seems unlikely Waymo has a permanent fix. If they did, they would tell NHTSA, who would not have needed to promote their investigation into a Recall Query.

Driving in closed areas of highway construction zones:

Many construction zones, especially on highways, have empty lanes that seem unused and obstruction-free, but are there either as buffer space or to reserve space for where work crews have not yet reached. Driving in such areas is against traffic rules, and presents an elevated risk to work crews.

The recall clearly did not fix the problem. California DMV determined that the vehicles pose an “imminent public safety hazard” and revoked highway driving permission. Waymo stood down highway operations in other states, which seems like a good idea for now. We’ll have to see what it takes to convince DMV that the fix really worked this time.

Interfering with emergency responders and related issues:

There have been ongoing problems with Waymo vehicles blocking streets when they shut down due to operational issues and remote assistants are unable to get them unstuck. This sometimes blocks emergency responders or otherwise interferes with their operations.

This is an open issue. The incidents keep coming. This might further break down into obstructing traffic, other interference with emergency responders, and putting unacceptable burdens on emergency responders to deal with robotaxi-specific issues. Expect more regulatory responses when we see more high-profile incidents.

Other potential patterns:

These problems have not resulted in big regulatory action. But they are problems nonetheless. Waymo needs to get ahead of these before regulators are forced to act.

(Dates are approximate and may correspond to a news article date rather than an incident date.)

The city of Austin provides a public dashboard of local robotaxi incidents. However, the incidents must be reported by the public. Waymo and other companies do not provide data about incidents they know about.

There are plenty of embarrassing Tesla videos as well, but the presence of a human supervising driving in most of those situations clouds the issue. So we leave those for another time. However, each company should learn from problems that happen to other companies. It seems that there are plenty of learning opportunities to go around.

Leave a Reply

Your email address will not be published. Required fields are marked *