Hacker News Reader

Comment by manzanarama | Hacker News Reader

I feel like you are injecting facts with a lot of assumptions. How do we know a human driver would have taken longer? What if the human would have seen the bike 200 yards up the road 45 seconds ago, and since bikes don't disappear, waited a half second before proceeding into the intersection. Or what if a small portion of the bike was visible?

It could have been absolutely unavoidable but I don't think we know that now.

View on HN

Let's be honest, anyone making any type of conclusion based off the information in the article is just giving in to their biases. None of us know enough to actually have any serious opinion on this specific collision.

I'm drawing the conclusion that this is a non-issue, or would be if autonomous vehicles hadn't got to be so political over the last year in San Fransisco. There are many people out for blood, they want a follow-up to the Cruise scandal.

One fact of the matter is that the cyclist wasn't seriously hurt, and Waymo has had many minor contact events in its 10 million+ miles of driving on public roads. We're hearing about this event because politics.

You are just proving my point. You are pro-autonomous vehicle so you are interpreting the few facts we know in way that benefits your side of the debate.

Referencing the cyclist not being seriously hurt is the most obvious example. It doesn't take much for a car to seriously injure a cyclist. It often just comes down to luck of the cyclists physical position and where they fall. Onto the hood of the car is safest, but they could have easily been caught under the vehicle, pushed into other traffic, or dangerously thrown down to the pavement. A cyclist walking away from this collision doesn't necessarily mean the next time a Waymo hits a cyclist will be just as safe. If there is some fault in the system that increases the odds of it hitting a cyclist (something that is impossible for us to know) it would be only a matter of time until an unlucky cyclist gets seriously hurt.

I'm not pro-autonomous vehicles. Under the hypothesis that they worked well enough that they proliferated widely and people came to rely on them, they would be the most enshittifiable service ever.

Also I ride a bicycle everywhere, I don't own a car. I've had many crashes while cycling, a handful of which involved another moving vehicle.

Speed is a good predictor of how much harm a moving vehicle can cause to a cyclist or pedestrian, and given that the Waymo was turning left off a 4-way stop, it couldn't have been going that fast, and even if it was, the waymo stopped soon enough to avoid doing serious damage. Maybe the cyclist veered into the Waymo, we don't know. Maybe we'll get video and then we can really pick it apart.

There is a battle going on right now between the Governor of California and SF city council over the city's inability to regulate the existence of autonomous vehicles on their streets after the fast proliferation of Cruise's AVs led to all kinds of traffic snarls and general irritation amongst the public.

Cruise has been operating in SF since 2019 and has had many incidents more severe than the one we're discussing now, but they got little attention because it wasn't so political then. Nowdays SF is looking for any excuse they can find to get AVs out of their city.

In legal terms I doubt the city will find what they need with this incident. With regards to public sentiment, the headline "Waymo hits cyclist in SF" is about as much as most people will read, and the details of the crash don't matter.

>I'm not pro-autonomous vehicles.

Maybe not, but you are clearly coming from an anti-anti-autonomous vehicle stance. That likely plays into why you are downplaying this collision despite admitting "we don't know" basic details about what happened here. You are able to recognize that this issue is politicized, but like everyone you are considering your own political bias as the neutral position when in actuality the neutral position here is to wait until we have more details on what actually happened.

I think at the very least we can interpret the situation as car hits cyclist in a blind spot and the cyclist was not seriously injured. Sources say there were 49,000 vehicle-cyclist injuries and 846 fatalities [1] and there were 3.2 trillion vehicle miles traveled [2] in 2019. So my math comes to around .015 injuries and .00003 fatalities per 1 million vehicle miles traveled. Waymo’s traveled 20 millions miles by 2020, so more by now but Im not finding a more recent number. This is the first time I’ve heard of a Waymo-cyclist injury (maybe it’s not; I’m not able to google around this new headline) and would put their injury rate at .05 per million miles significantly higher than the other statistic but to get away with minor scratches is nonnegligible since it’s easy for injuries to be worse. At 20 million miles Waymos this rate doesn’t make them look good. I want to be optimistic about their injury rate getting better as they program around blind spots safer.

[1]https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/...

[2]https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/...

This is the same type of math that Tesla uses to suggest that their Autopilot is safer than humans, but the problem with this approach is that not all miles driven are equal. I guarantee that the injury rate per mile is much higher on the road in which this collision took place than the 0.015 national average. There are obviously roads in which collisions with cyclists are incredibly rare like interstate highways in which a cyclist even using the road is rare. That means there needs to be roads which greatly exceed the national average and those are often city streets with heavy cyclist traffic like this one where the collision occurred.

> We're hearing about this event because politics.

I think reporting on integrating autonomous driving into roads is a news worthy subject even outside of the agenda of a particular publisher. The good things are reported for autonomous driving as well. I don't live in SF but know that Waymo and Cruise both had fully autonomous vehicles in SF.

Having access to a comprehensive database of incidents and near misses would be informative. A single incident where only incomplete information is available doesn't tell us much.

> Having access to a comprehensive database of incidents and near misses would be informative. A single incident where only incomplete information is available doesn't tell us much.

Agreed!

I wasn't commenting from a safety perspective, but from a news perspective. Recently Cruise has had regulatory action taken against them from the California Department of Motor Vehicles due to an autonomous vehicle accident. Waymo, another company working on autonomous vehicles also has an accident! Sounds news worthy to me.

Of course reporting on a crash will always have some negative connotation for Waymo, and I hope the regulators look at more than individual incidents to evaluate the safety of Waymo's autonomous vehicles. I did learn that Waymo recently had an accident in a time period of scrutiny for autonomous vehicles as they further integrate into roads.

When Waymo was known as the Google Self-Driving car project they were cavalier about safety, but became much more conservative after spinning out as Waymo under John Krafcik in 2017.

Waymo has not had any serious incidents and these days it seems they're doing what they can to remain low-key and avoid attracting negative attention to themselves. Like you said, when Cruise, Uber or Tesla behave recklessly, it can't help but bode poorly upon Waymo in the eyes of the public.

We can't directly compare what these companies have going on under the hood because it's all quite proprietary. Waymo nonetheless has been chipping away at the problem for longer and with more resources at their disposal than any competitor. Waymo's 'Driver' is far and away the most experienced. While I'm fully confident making that claim, there's no easy way to measure it or make an emprirical comparison to other drivers.

If you want to play this game and you aren't very experienced, you can fake it by being reckless. You can make it seem to investors that you're better than you are by putting hundreds of vehicles on the road. Investors want results. You have to be able to point to a line on a graph that goes up and to the right and say "look at all these new benchmarks we hit! More cars! More miles!"

Waymo is effectively patronized and will run at a loss for as long as they need to without any pressure to fake it until they either make it or break it. It's Larry and Sergey's pet project. It's the one they won't let go of. A single scandal can really mess things up.

This feels like the "no vehicles in the park" all over again.[0] We're all naturally inclined to have assumptions, but I think many are not aware of this. I think all have the capacity, just not the habit.

[0] https://news.ycombinator.com/item?id=36453856

I've been fairly impressed at Waymo's defensive driving skills, down to it being able to tell when someone turns around towards the road with an eye to cross, & the car slowing-down/giving a wider berth.

We will need to wait and see (i.e. footage) for more facts on this one, but I would caution thinking that Waymo's don't have object permanence capabilities.

(Usual disclaimer around 1st-hand experience etc etc)

What we do know is that this incident will be analyzed with the level of attention we would give to a plane crash. If there was any way the incident could have been avoided, the engineers will probably figure it out. It’s likely that software will be updated to implement a fix as well, and this type of accident will become much less likely across Waymo’s entire fleet. That seems like a really good result.

Only the driving software will be updated, the intersection will not be redesigned to reduce the severity of driver mistakes. It’s not the best outcome, ideally someone will analyze the case and produce the changes that needed to made to the driving software, the intersection, and biking behavior (aka the Dutch approach).

Fair, but then again, this would be taking drastic measures for a single data point. A full redesign would be quite a stretch for a single accident and nearly impossible to scale. The least you'd need to do is check for a pattern of accidents.

Yet that’s exactly what the Dutch do:

https://www.bicyclelaw.com/after-every-crash-in-netherlands-...

> In the Netherlands, accidents like these are followed by intense investigations, street redesign, and criminal prosecution on a level wholly different from Boston, where a slew of bike fatalities in recent years have prompted modest on-street changes and police crackdowns on bicyclists running red lights. But there have been few street design overhauls and no criminal convictions of motorists in those fatal accidents.

I’m pretty sure there is an HN discussion on this article somewhere.

Ah, the highest level of scrutiny is limited to and mandatory for fatalities, which makes sense.

Journalists swarm all over it. Waymo engineers will take their learnings. But honest question: is there an official, scrutinous authority also coming into this that would be able to hold Waymo to account? Maybe like the NTSB?

The thing is, human drivers do not and will never have such scrutiny. In this case, there is some hope that it will happen before these cars are widely deployed, even if said authority is the court of public opinion.

In my experience with motorcycling and bicycling as a primary mode of transport I'm happy if the human drivers attempt to stop at all

Cyclists are supposed to stop at stop signs, if not stop at least yield. At least in this case (if true), the cyclist followed the truck through a 4 way intersection (assuming it's a 4-way stop), that would put them in the wrong.

Hope waymo releases the footage.

There’s rules as written and then rules as followed — I’d much prefer a system that recognizes the rules that people tend to follow/bend/break — as a cyclist I too will often “convoy” with a bigger vehicle as it provides some additional protection most of the time (though obviously not here)

> rules as written/rules as followed

As a bicyclist in San Francisco, if I follow the rules as written, I cause traffic. Cars expect me to blow through four-way stop-sign intersections, and if I stop and wait for the cars, the drivers get confused & don't want to go (afraid of hitting me, I suspect).

In terms of right of way, the rules as followed seem to be pedestrian > bike > car.

I sometimes give way to bicyclists at four-way stops because I can't be sure they didn't get there before I did and I didn't see them because they're small.

And of course, if a vehicle, be it a bicyclist or car, enters an intersection when I have right-of-way, it's not like I'm going to start crossing and intentionally run into them.

Where bicyclists really risk their lives if they start assuming cars will give way to them is when they blow through two-way stops, especially at night or at one of the many intersections with poor visibility.

So are cars and yet it's common practice among drivers to simply slow down to 5-10 mph, declare that a "stop" and proceed into the intersection.

Note that there are bike lanes on the likely street where this occurred. The cyclist was not necessarily following the truck but could have been parallel (and overtaken).

As a cyclist, I'm invisible to drivers quite often even when I'm not hiding behind a truck. Many drivers also immediately forget you after overtaking you just seconds ago and attempt a right-hook. It is of course possible that a human's advanced reasoning could have avoided the situation as described, but imo that would've been sheer luck or a very unusual driver.

Reply on news.ycombinator.com