The Waymo's were just going really slow through the intersection. It seemed that the "light is out means 4-way stop" dynamic caused them to go into ultra-timid mode. And of course the human drivers did the typical slow and roll, with decent interleaving.
The result was that each Waymo took about 4x as long to get through the intersections. I saw one Waymo get bluffed out of its driving slot by cross traffic for perhaps 8 slots.
This was coupled with the fact that the Waymos seemed to all be following the same route. I saw a line of about a dozen trying to turn left, which is the trickiest thing to navigate.
And of course I saw one driver get pissed off and drive around a Waymo that was advancing slowly, with the predictable result that the Waymo stopped and lost three more slots through the intersection.
On normal days, Waymos are much better at the 4-way stops than they used to be a few years back, by which I mean they are no longer dangerously timid. The Zoox (Amazon) cars are more like the Waymos used to be.
I expect there will be some software tweaks that will improve this situation, both routing around self-induced congestion and reading and crossing streets with dead lights.
Note that I didn't see any actually dead Waymos as others have reported here. I believe this is an extreme failsafe mode, and perhaps related to just too much weirdness for the software to handle.
It would be interesting to see the internal post mortem.
Is there a name for this (and related) effects? Obviously, in a group of several hundred thousand people, there will always be at least a few people that complain about something for the exact opposite reasons. That's not a signal of usefulness. I feel we need a name for the some-rando-has-an-opinion-that-gets-picked-up-and-amplified-by-"the algorithm" phenomena. And the more fringe/out-there, the more passionate that particular person is likely to be about this issue, when "most" people feel "eh" about the whole thing.
when a waymo can get a traffic ticket (commensurate with google's ability to pay, a la the new income-based speeding ticket pilot programs in LA and SF), and when corporate officers down to engineers bear responsibility for failures, i think a lot more people will stop seeing these encroachments onto our commons as a nuisance.
story time: i've literally had one of those god awful food delivery robots run straight into me on a sidewalk. once, one of them stopped in my way and would not move, so i physically moved it myself and it followed me to my apartment. i'm about to start cow-tipping them (gently, because i don't want a lawsuit alleging property damage, even though they're practically just abandoned tech scrap without a human operator nearby to take responsibility).
And if you’re Waymo, it’s a short-term reputation hit but great experience to learn from and improve.
Just because no integer lives were wasted doesn't mean we can't sum the man hours and get a number greater than 1
Okay, let's see if they actually do it this time.
What post mortem? The whole fleet reverted to it's "baseline" of acting like a hysterical teenager on day 1 of driver's ed. Obviously there's serious collateral damage to overall system performance when you just create thousands of those people out of thin air but that's "other people's problem" as far as waymo is concerned.
In addition, there are 4-way stop signs all over SF and tourists regularly comment on how they work here.
The law is clear - yield to the right, but that is a pretty slow system in congested roads.
The local custom in SF is that someone is usually obviously first, rightmost, or just most aggressive, and opposing pairs of cars go simultaneously, while being wary about left turns.
Of course pedestrians have right of way in California, so someone in a crosswalk gives implied right of way to the road parallel to the person's crosswalk.
The result is 2x or better throughput, and lots of confused tourists.
So ... with the lights out on a Saturday before Xmas, there was a mess of SF local driving protocol, irritated shoppers, people coming to SF for Xmas parties, and just normal Saturday car and foot traffic.
I thought Waymo did pretty well, but as I said, I didn't see any ones that were dead in the middle of the street..
Why are you saying they got pissed off? Going around another vehicle that is blocking the road sounds like basic driving to me.
One of the biggest issues with current state of tech I see is, where these cars usually are. They're in cities, and most often in very dense ones, and ones in the south. These are effectively perfect conditions.
From my perspective, I wonder how these cars will behave with ice on the road, with snow, or a typical Montreal Wednesday of "It's a blizzard, you can't see 10 feet, there is snow on the road and ice, it's slippery, all the lines and street markings are obscured completely, oh and the power is out and there are no traffic lights."
Some of this can be resolved by snow tires, or even studded tires which are legal in Quebec. It should be noted that Quebec plows the roads less, and uses less dirt and salt on the road, and also enforces a law that snow tires are on cars in the winter. Of course studded tires give insane grip on ice, but have reduced grip on rain.
And it can 10C and rain, then freeze, then be a blizzard, then move to -40C, all in a few days.
But anyhow, my point is if a Waymo is slow with a missing traffic light, how will it act with a missing traffic light, and 10ft visual range of reflective snow in the air, no ability to see lines on the street, and so on. Humans are great at peering and seeing mostly obscured indications of an intersection, but this is still challenging for a car with a top priority of safety.
Here's another example. The cameras in my car are constantly obscured by slush, dirt, and such on the windshield and all over the car. All the roads are coated with dirt to help with slipping on ice. I often have my car absurdly complaining that cameras are covered, and there's no assist this and that, just because the entire car is coated in dirt.
How will a Waymo operate with all sensors covered in dirt?
There are probably solutions. But it feels like it will be a long while before such cars treat a normal day in winter, as usual.
It should be noted that I've simply discussed downtown Montreal. What of a rural area? And by rural, I mean houses 1 km apart, also with a blizzard, all lines obscured on the road, and meanwhile Canadians just intuitively know where and how to drive it. We just slow down a bit (from 120km/hr to maybe 70km/hr) and just drive on our merry way. If we try to stop, distances are greatly extended, and of course in some places without care you'll just slide into the ditch.
Of course that's just a Wednesday, and you can read the 'signs of the road', and sort of tell where to slow down more. Where to take more care.
Sometimes, you'll see a bunch of cars in the ditch, and think 'Ah, must be particularly slippery here', and slow down a bit more.
“Thought of the day, and I wish there were a way to get this to legislators:
Come the next Big One earthquake, all of San Francisco’s emergency services will be blocked by Waymos.”
I’m AMAZED they’re not designed to handle this better. This does indeed seem like a massive problem. “Oops we give up” right when things get the worst? How is this OK?
I’ve been very impressed by Waymo’s more cautious approach. Perhaps they haven’t fully thought through the ramifications of it though.
There is a chorus of voices here on HN that have tried to do this openly, obviously, myself included. It seems to be almost universally derided by people who apparently assume that we're just trying to hurt a start up out of anti-environmental sentiment and jealousy.
There are more ways to get "self-driving cars" wrong than there are to get it right. Driving is far more complex than the hackers here on Hacker News seem to want to concede, and even if that wasn't the case, I'm not sure where the sentiment that a trillion dollar corporation is naturally going to implement this system with the best interests of society in hand comes from.
It's a genuine frustration here.
I believe the answer is far more complicated than it seems and in practice having the cars stay still might have been the safest option any of the parties could agree on (Waymo's office, the city traffic people, state regulators, etc).
There are people thinking this stuff out and those cars can 100% pull over automatically but an explicit choice was made not to do so for safety.
Maybe I'm reading things wrong, but it sounds like the top comment wants waymo to be better, and you want waymo to be off the roads. You're not talking about the same kind of "thinking through the ramifications".
The sentiment comes from the corporation itself. With this much money at stake you know they have a hand in steering the conversation and that includes on sites such as this.
I feel like self-driving cars are, pretty objectively, the single least environmentally friendly mass transit solution (more cars being made and using more rare-earth minerals to produce them, more cars being driven rather than increasing public transit usage). What's the argument that not liking self-driving cars is "anti-environmental"?
Decades ago I recall talking to a fireman expressing a question of what happened if there was a car blocking their access in an emergency and he made it clear that the bumper on the front of the truck and the truck's healthy diesel engine would usually take care of the problem very quickly.
This has been the MO for "tech companies" for the past 20 years. Meanwhile I'm told I'm paranoid when the industry of "move fast or break things" decides to move into mission/safety critical industries and use its massive wealth to lobby for deregulation to maintain its habits.
We certainly have BS regulations done to constrain competition. But I'd wager a good 80% of them exist for good reason.
Not because a bunch of cars that are perfectly capable of moving are just sitting there blocking things purposefully waiting for the driver in the sky to take over.
And what if, due to $BIG_DISASTER they won’t be able to for a week?
Were any emergency vehicles actually blocked?
We have an actual failure here–step one is identifying actual failures so we can distinguish what really happened from what hypothetically could.
They need to drive or pull over. Never just stop there in the road and wait.
The basic thing is to treat everything like a four-way stop sign.
And I believe Waymo remote access only allows providing high level instructions (like pull over, take the next right, go around this car, etc) precisely because full direct control with a highly and variably latent system is very hard/dangerous.
And in an emergency situation you’re likely to have terrible connectivity AND high level commands are unlikely to be sufficient for the complexity of the situation.
Perhaps in such cases they can pull over in a safe place, or if they have an occupant ask them if they wish to continue the journey or stop.
Perhaps they already do this, I have no experience with autonomous vehicles.
Presumably Waymo will make sure they can handle this situation in the future, but I'm not sure there's a really satisfactory solution. The way you're supposed to handle an intersection with no lights (treat it as a stop sign intersection) doesn't work very well when no one else is behaving that way.
There was a lot of confusion, and some people took advantage of it to rush through, but for the most part it was pretty orderly. Which makes sense because in many parts of the world where there are no traffic lights or stop signs, people get on just fine.
The Waymo’s, on the other hand, were dropping like flies. While walking from Lower to Upper Haight I spotted a broken Waymo every handful of blocks. The corner of Haight & Fillmore was particularly bad, with 3 of them blocking traffic in both directions — in the path of both the 7 and 22 bus lines.
Well, sort of. Road injuries / fatalities in countries without these kinds of regulations are about an 3-4x higher than in those that do have them.
wow, cascading failures. I'll bet this is the tip of the iceberg.
Herd mentality also helps here. We see the first few people do this and we will follow along pretty quickly.
My own experience has given me a somewhat more-nuanced take.
At first, it's akin to the path of evil. Way too many people just zoom through intersections with dark traffic lights like they're cruising unimpeded down the Interstate, obvious to their surroundings. Some people get grumpy and lay on the horn as if to motivate those in front of them to fly through themselves.
But many people do stop, observe, and proceed when it is both appropriate and safe.
After awhile, it calms down substantially. The local municipality rounds up enough stop signs to plant in the middle of the intersections that people seem to actually be learning what to do (as unlikely as that sounds).
By day 2 or 3, it's still somewhat chaotic -- but it seems "safe" in that the majority of the people understand what to do (it's just stop sign -- it may be a stop sign at an amazingly-complex intersection, but it's still just a stop sign) and the flyers are infrequent-enough to look out for.
By day 5 or 6, traffic flows more-or-less fine and it feels like the traffic lights were never necessary to begin with. People stop. They take turns. They use their turn signals like their lives depend on it. And the flyers apparently have flown off to somewhere else. It seems impossible to behold, but I've seen it.
But SF's outage seems likely to be a lot shorter than that timeline, and I definitely agree with Waymo taking the cautious route.
(but I also see reports that they just left these cars in the middle of the road. That's NFG.)
In absence of priority roads there is also the "right before left" rule which means that the car coming from the right if they would conflict in time is the car that has priority. It's also always illegal to enter an intersection if you can't immediately clear it; that seems to work better when there are no green traffic lights to suggest an explicit allowance to drive, though.
In contrast many years ago I lived at an intersection that had almost no pedestrians back then and a few times for a power outage limited to our building and that intersection. I enjoyed standing on my balcony and watch traffic. It mostly worked well. Cars did treat it like a intersection with stop signs. There two issues happened though. One was when there was no car already stopped and about 10%-20% of drivers didn't realize there was an intersection with lights out and just raced through it. The other ironically were bicyclists. 90% of the just totally ignored there was an intersection. That was especially scary when they arrived at the same time as one of those cars who didn't realize it either.
I’ve long been curious if people in S.F. who are used to Waymo “behavior” – including myself – behave differently when a Waymo is involved. For the most part, Waymos are extremely predictable and if you’re on the road as a pedestrian, cyclist, or car, are you more aggressive and willing to assert your turn in the road? Curious if Waymo has done a psychological study of how we start to think about the vehicles. I know many runners, for example, that’ll stride right into the crosswalk in front of a Waymo but wouldn’t dare to do that in front of other cars. Similarly, anecdotally people treat public transit vehicles differently than a civilian car: folks are more willing to let them drive even if out of natural turn.
One interesting effect is that there are also often pedestrian crossings that have priority over everyone. Normally those are limited by lights, but without lights a steady stream of pedestrians stops all traffic. Seen that happen in Utrecht near the train station recently, unlimited pedestrians and bikes, so traffic got completely stuck until the police showed up.
My purely anectodotal experience is that the response is variable and culturally dependent. Americans tend to treat any intersections with a downed stoplight as a multi-way stop. It's slow but people get through. I've experienced other countries where drivers just proceed into the intersection and honk at each other. (Names withheld to protect the innocent.)
It seems a bit like the Marshmallow test but measures collaboration. [0]
[0] https://en.wikipedia.org/wiki/Stanford_marshmallow_experimen...
> Traffic Light Not Working
When a traffic light is not working, stop as if the intersection is controlled by STOP signs in all directions. Then proceed cautiously when it is safe to do so.
https://www.dmv.ca.gov/portal/handbook/california-driver-han...
But those complex multiple lanes in all directions + turn lanes...
They do break down. I think they are a breeding ground for confusion and frustration.
Actually there are times of day when I find it preferable that the traffic lights are down.
In Ireland, there are precisely two passenger routes still operated with locomotives, and there’s a tender offer out to replace one of them with a (really wacky; diesel, battery, _and_ overhead lines in two voltages!) multiple unit.
When you link the cars together, they usually switch to a hub that's a 10-15 minute walk from your destination instead of your destination and the compartments are occasionally shared with unstable and violent people, which while possibly "efficient" in some metrics, are downsides that many people would rather avoid. Personal compartments are a real differentiating advantage.
A quaintly American complaint. A 10 minute walk being an issue is very a learned helplessness my fellow Americans suffer from.
But unfortunately the 10-15 min walk is only possible in a couple cities. most Americans day to day experience of public transit is spaced out buses that don’t work well for single family sprawl and strip malls parking lots where walking is treated as undesirable. Car oriented rather than people oriented urban planning (or lack thereof) is the original cause.
Buses in Paris run with IIRC 60kWh battery and pantograph charger at every other station. Packs (not cells) recently dropped to below $100 kWh. At $6k thats probably what city pays for couple of replacement seats (gold plating et all).
Streetcars?! In San Francsico?!?
Better leave now for you doctor’s appointment tomorrow and I hope you scheduled three days off from work.
Public transport for things like metro/trains/trams/buses are honestly underrated.
But even without them getting better, as far as I know there were zero waymo fatalities due to this.
That's more than I can say about Helene, where there was at least one fatality due to traffic light outages.
Lets not forget that a big part of why we want Waymo is that it has already lead to a dramatic decrease in fatal accidents. They are a great company that will do a lot of good for the world. One bad night (in which noone was hurt, in part because of their cautiuosness) shouldn't negate that.
I was curious if Waymo has even been involved with a crash that killed someone, so I looked it up. The answer is yes - there was a Tesla going 98mph in SoMa whose driver died after hitting a Waymo. Clearly we should shut down Waymo until they can handle that situation!
Even if every US city had Waymos blocking the street for every single disaster, as they did here. I find it extremely unlikely that even the indirect deaths would come close to that number. And that's assuming Waymo learn from this lesson. Which they will.
[1] https://www.nhtsa.gov/press-releases/nhtsa-estimates-39345-t... [2] https://storage.googleapis.com/waymo-uploads/files/documents...
A human can combine a ton of context clues. Like, "Well, we just had a storm, and it was really windy, and the office buildings are all dark, and that Exxon sign is normally lit up but not right now, and everything seems oddly quiet. Evidently, a power outage is the reason I don't see the traffic light lit up. Also other drivers are going through the intersection one by one, as if they think the light is not working."
It's not enough to just analyze the camera data and see neither green nor yellow nor red. Other things can cause that, like a burned out bulb, a sensor hardware problem, a visual obstruction (bird on a utility cable), or one of those louvers that makes the traffic light visible only from certain specific angles.
Since the rules are different depending on whether the light is functioning or not, you really need to know the answer, but it seems hard to be confident. And you probably want to err on the side of the most common situation, which is that the lights are working.
My approach was to get closer into the intersection slowly and judge whether the perpendicular traffic would slow down and also try to figure out what was going on or if they would just zip through like if they had green.
It required some attention and some judgement. It definitely wasn't the normal day to day driving where you don't quite think consciously what you're doing.
I understand that individual autonomous vehicles cannot be expected to be given the responsibility to make such a call and the safest thing to do for them is to have them stop.
But I assumed there were still many human operators that would oversee the fleet and they could make the call that the traffic lights are all off
Not sure what about this isn’t funny. Nobody died. And the notion that traffic lights going down would not have otherwise caused congestion seems silly.
I am pretty sure Waymo does not disclose how many human interventions they get. It would destroy their magic aura. A fancy RC car with self-driving experimental features is not very futuristic after all. By all the evidence, that’s what we observed when the internet went out. I don’t buy the 4-way stop explanation. Waymos handle 4-way stops just fine on an average day. I drive alongside them daily.
I’ve long suspected that they get many human interventions on the road, frequent enough that when the internet connectivity slowed down to a crawl across the city, Waymos could not get themselves unstuck from a variety of situations and simply just blocked the roads. That’s not a paragon of safety, nor is it “self-driving”. Self-driving cars were 10 years away in 2015, and in 2025, they are still 10 years away.
Makes me think there are likely other obvious use cases they haven’t thought about proactively either.
We have zero evidence a power outage wasn't foreseen. This looks like a more complex multi-system failure.
Once you’re on public roads, you need to ALWAYS fail-safe. And that means not blocking the road/intersections when something unexpected happens.
If you can physically get out of the way, you need to. Period.
Certainly a better way to handle this would have been to pull over. I think stopping where ever it happened to be is only acceptable if the majority of sensors fail for some reason
I being the hero I was, wanted to keep the show running, bought some candles, ovens worked fine, water worked fine (for now). I wanted to charge cash. But eventually big boss came and shut us down since light wasn't coming.
And he was right, cooking and working under those conditions is dangerous for the staff, but also for the clients, without light you cannot see the food, cannot inspect its state, whether stale, with visible fungi, etc...
Yes, the perfect worker would still operate under those conditions, but we are not perfect, and admitting that we only can provide 2 or 3 nines, and recognizing where we are in that 0.01% moment, is what keeps us from actually failing so catastrophically that we undo all of the progress and benefits that the last bit of availability would have allowed us.
AIUI, it was the irregularity of the uncontrolled intersections combining with the “novel” (from the POV of the software) driving style of the humans. In dense areas during outages signaled intersections don’t actually degrade to 4 way stops, drivers act pretty poorly.
The normal order and flow of traffic broke down. The software determined it was now outside its safe parameters and halted.
Certainly not ideal, and the should be a very strong regulatory response (the gov should have shut them down), and meaningful financial penalties (at least for repeat incidents).
And my question is why did it halt instead of pull over?
Perhaps traffic lights being out is what caused the cars to stop operating autonomously and try to phone home for help, or perhaps losing the connection home is itself enough to trigger a fail safe shutdown mode ?
It reminds a bit of the recent TeslaBot video, another of their teleoperated stunts, where we see the bot appearing to remove a headset with both hands that it wasn't wearing (but that it's remote operator was), then fall over backwards "dead" as the remote operator evidentially clocked off his shift or went for a bathroom break.
Things go wrong -> get human help
Human not available -> just block the road???
How is there not a very basic “pull over and wait” final fallback.
I can get staying put if the car thinks it hit someone or ran over something. But in a situation like this where the problem is fully external it should fall back to “park myself” mode.
https://waymo.com/blog/2025/12/autonomously-navigating-the-r...
"Navigating an event of this magnitude presented a unique challenge for autonomous technology. While the Waymo Driver is designed to handle dark traffic signals as four-way stops, it may occasionally request a confirmation check to ensure it makes the safest choice. While we successfully traversed more than 7,000 dark signals on Saturday, the outage created a concentrated spike in these requests. This created a backlog that, in some cases, led to response delays contributing to congestion on already-overwhelmed streets."
1. Nobody at Waymo thought of this,
2. Somebody did think of it but it wasn't considered important enough to prioritize, or
3. They tried to prep the cars for this and yet they nonetheless failed so badly
Also I'm not sure it's entirely AI's fault. What do you do when you realistically have to break some rules? Like here, I assume you'd have to cut someone off if you don't want to wait forever. Who's gonna build a car that breaks rules sometimes, and what regulator will approve it?
People get promoted for running DiTR exercises and addressing the issues that are exposed.
Of course the problem is that you can't DiRT all the various Black Swans.
PG&E outages in S.F. leave 130k without electricity
My personal opinion. With number of cars I saw flying through blacked out intersection -- major intersections -- I'm very happy that Waymo had a fail safe protocol for such a "white swan"-style event (that is extremely rare, but known-to-happen event).
I saw a damn Muni bus blow through an minor intersection, and was just shaking my head. So many dumbasses behind the wheel, it's miracle no one was killed, and everyone seems to be concerned with "the flow of traffic."
Perhaps we are just more used to traffic lights being off/broken (and we are, as this is, anecdotally, more like a weekly occurrence at some point during your trip to work, for instance)?
Is this a violation of the California Vehicle Code? Generally it seems to disallow non-emergency vehicles from traveling with blinking lights except for turn signals (and brake lights responding to a braking action).
- New Orleans taxi cab driver
- Houston gang banger
- Los Angeles traffic weaver
and most importantly,
- Saudi Arabia Toyota Camry driver and 360 drift hobbiest (with bonus 2 wheel tire change)
Each city will have its own nuances.
Why don't the regulators publish the list?
Well there's your problem.
If they truly rely on teleoperation, that's at least 20ms in the best case, and can grow a lot with interference.
I always assumed these things have some autonomy.
The Waymo video has over a dozen cars, at least 6 pedestrians crossing streets (many more on the sidewalks), and is a 5-way intersection.
These are cherry picked examples. Either advertising or propaganda.
I think the emergency "phone home" protocol requires a phone, presumably with enough channel capacity for reasonable video feeds. I wouldn't be surprised if the dead in the road Waymos were lacking connectivity.
There is of course also a possibility that the total demand exceeded the number of people at Waymos available for human intervention.
Though the safety driver disengaged twice to let emergency vehicles pass safely.
https://www.tesla.com/ownersmanual/modely/en_eu/GUID-A701F7D...
Looks like it treats it as a 4-way stop. Is this because Tesla has more training data?
Don’t think I have had a totally inactive light. I have had the power is out but emergency battery turned to blinking red light, and it correctly treats as a stop sign.
Its human takes over. FSD is still Level 3.
(Robotaxi, Tesla's Level 4 product, is still in beta. Based on reports, its humans had to intervene.)
Waymo's problem is obvious in hindsight, and quite embarrassing for them, but it can be solved with software improvements. Tesla's FSD already treats dark traffic lights as stop signs, so I would bet on Waymo fixing this as soon as they can.
But transportation that depends on infrastructure along the whole route (such as trains and busses powered by overhead lines) are always going to fail in these situations. I think that's acceptable considering how rare these events are.
Sure, I go out and drive around on the roads for no reason all the time. I'll avoid doing that during the crisis.
Or look at the traffic and decide if you REALLY want to spend an hour or more in gridlock for whatever activity you are considering.
And maybe wherever you wanted to go is closed because they don't have power either.
It is a perfectly reasonable request.
The fact that you acknowledge that is was a "crisis" implies pretty strongly that you understand that a priority evaluation might be useful.
Some folks enjoy driving.
SF is pretty brave (stupid?) for allowing itself to be a beta test for self driving.
