Autonomous vehicle drove through police tape at a crime scene in San Francisco | Hacker News Reader

Autonomous vehicle drove through police tape at a crime scene in San Francisco

twitter.com

37 pointsperihelions42 comments

Open articleView on HN

Every time something like this makes the news, I would like to ask:

1. How many total vehicle-miles have this level of autonomous vehicles driven in all of history?

2. How many times has an incident like this happened?

3. Now what's that ratio for humans.

Truth of the matter is, if the event was "One [bad] driver drove through police tape" this wouldn't be a story.

I know even single-events are illustrative without a denominator, but it's hard to think about how much to fault autonomous vehicles as a whole without that.

Every time someone posts this kind of response i wonder who is the apologist for the human who does something like this because normally functioning people never would, you have a robot that being hailed for its driving perfection acting like a drunk or a psychopath and people trying to convince you it’s not so bad because it’s brothers statistically don’t do it that often.

I'd just like to point out one substantial difference. The autonomous part of the car is software so when if fails in a strange way that is also dangerous the natural question is how much is this going to happen across all the cars that have this software, and also how much can engineers understand why this happened to prevent it in the future. The answers to those questions are not straight forward so even a single incident has much wider implications than a human making a mistake.

>if fails in a strange way that is also dangerous the natural question is how much is this going to happen across all the cars that have this software

If an AI messes up, it might do it again. If a human driver messes up, another human driver might do it again. I'm not seeing a substantial difference.

Humans aren't all running the same software of course, but that just introduces some [additional] probabilistic uncertainty to our estimate. Fundamentally the situation with human drivers remains the same (or worse). In fact, we can be 100% certain that the 'human software' running on all the other cars on the road is not bug-free!

> and also how much can engineers understand why this happened to prevent it in the future

This is legitimately a substantial difference.

Yet this difference is an advantage for autonomous cars, while the reaction to these headlines is almost universally negative toward the technology.

> In fact, we can be 100% certain that the 'human software' running on all the other cars on the road is not bug-free!

The roads, traffic system, laws and institutions etc., have all been designed for humans, and have been constantly updated and improved for many decades. This means autonomous vehicles might make a mistake that is practically impossible for a human to make. As an exaggerated example to make the point, it could malfunction in the middle of a highway and break full stop, or maybe keep going straight and fly off the highway unto a building. That said, they will get safer just like the human system, and probably way more safer eventually.

>This means autonomous vehicles might make a mistake that is practically impossible for a human to make.

Exactly—the mistakes are newsworthy because they're unusual (both in specifics and the general novelty of "AI did X"), and not because (exaggerations aside) they're categorically any worse than the mistakes humans make.

We are 100% guaranteed that humans aren't perfect drivers. A headline where we "discover" an AI is imperfect, therefore, doesn't let us conclude AI is worse than human drivers.

On the other hand, autonomous systems can benefit from direct learning transfer through shared software, whereas humans cannot. In this respect AI is categorically superior to human drivers.

Then don't sell autonomous driving as anything other than "we hope it probably won't screw up more than any other bad driver" and fully admit that the autonomous driving companies are conducting a giant beta (maybe alpha) test using real roads and real people. This is exactly the sort of things those companies have been promising can't happen.

None of these cars should be beta testing on the public road. Everyone knows of Tesla's woes, but I remember seeing on YouTube a Waymo car not being able to proceed past an orange traffic cone:

https://www.youtube.com/watch?v=B4O9QfUE5uI

All of them have absolutely poor implementations of "AI" and should not be trusted with human life.

I mean Tesla and Waymo are hardly similar, just like failing to proceed past an orange cone is very different from putting someone's like at risk.

They have to be tested on public roads at some point, don’t they?

Tesla’s approach seem many tiers more reckless than Cruise/Waymo to me.

Can’t they at least hire a driver to sit in the car?

They have/do! And and at some point they have to remove them and they’re probably still going to encounter situations they don’t handle perfectly.

Seems to me these cars just need a way for someone outside of the car to halt the car and call for a remote driver and open up communications.

This whole snafu is just because different parts of transportation are lagging.

The whole technology needs some integration. No-go zones are already available for robo-vacuums. Maybe that crime tape needs some rfid tags in it or something.

And why don't cars have an RFID VIN tag? So I could open an account and just drive into my parking garage, drive out again without stopping? Or park in any outdoor city space and be charged similarly?

My car wash for heaven's sake has this. But I have to stick a doohickey on the inside of the windshield when I register my contract.

The solutions are mostly simple. It's only waiting for regulation/integration.

> Maybe that crime tape needs some rfid tags in it or something.

Or maybe, tech companies shouldn’t be testing tech incapable of following laws on public roads.

> The solutions are mostly simple. It's only waiting for regulation/integration.

Oh for fucks sake, the solution is already as simple as it can get; don't cross a police line. This is already a solved problem.

> the solution is already as simple as it can get; don't cross a police line.

"Why the fuck can't Linux just find all the pictures I have of my dog? Who the fuck doesn't know what a dog is? Are these programmers morons?"

Later:

"Just send an email every Wednesday at 3pm. How fucking hard can it be?"

(Wikipedia: "Leap seconds are irregularly spaced because the Earth's rotation speed changes irregularly. Indeed, the Earth's rotation is quite unpredictable in the long term, which explains why leap seconds are announced only six months in advance. [...] Other contributing factors are the movement of the Earth's crust relative to its core, changes in mantle convection, and any other events or processes that cause a significant redistribution of mass. [...] For example, glacial rebound shortens the solar day by 0.6 ms/century and the 2004 Indian Ocean earthquake is thought to have shortened it by 2.68 microseconds.")

"I don't care, Hennimore, just do the fucking Wednesday thing. And make Linux get my photos of Cupcake."

Or maybe they could task some HI¹ with collecting a comprehensive list of different scenarios which rarely occur in the streets and in which a self driving vehicle is expected to perfom well: accidents, making space for emergency vehicles in stuck traffic, tunnel fires, ...

And then when they tested against all that for a decade on their closed test tracks they can let the thing onto the streets.

Because even if there were such block out zones, and there was a way to prevent abuse of that system, you'd still have to handle situations without those.

¹ HI: Human Intelligence

That would be the other end of the scale I guess, being ultra-careful.

But in the US at least, we let teenagers drive after maybe a few hours practice. They make mistakes all the time, and we tolerate them.

So our comfort zone is somewhere in between. Perhaps a 'UA Listing' that has a standard for performance in certain scenarios.

The pic was taken less than 5 mins' walk from my home.

AFAIK from local news reports, police were involved in a standoff with someone in a residence above a grocery store. That person was shot and killed. None of the news reports mentioned any crime being committed on the roadway.

https://missionlocal.org/2023/05/police-shooting-bosworth/

Same here: https://news.ycombinator.com/item?id=32828898#32833 (no one noticed).

This means that governments need to have a system of updating the public about any action, diversion on public infra as fast as possible

No, that was ethically conscious vehicle surrendering to the police for the crime committed. Now vehicles needs some compassion.

Do the autonomous vehicles have an Emergency-stop button somewhere?

So who gets cited for interfering with a crime scene in this case?

In a perfect world the company which operates those vehicles, represented by the CEO (see VW).

Yes, but I also think the driver has some responsibility. So they should both pay for it.

The tweet doesn't contain many details, but an "autonomous vehicle" in that area could actually mean no driver.

Even Autopilot, the most advanced autonomous vehicle technology currently available to consumers, requires consistent driver attention at all times.

And various non-consumers operate driverless cars in SF.

You can go take a Cruise ride in SF right now with no human being in the front seats at all.

Depends on your definition of “currently available to consumers” but no, Autopilot is not ahead of everyone else.

Cruise being available to a small group of testers in one city in the USA is not "generally available to consumers."

Sneaking in more qualifiers are we? :)

My claim: Tesla Autopilot is the most advanced autonomous tech currently available to consumers.

Your claim: Cruise is the most advanced tech currently available to consumers.

My refutation: Cruise is only available in one city in the USA.

Are you braindead? Why is there so much blind anti-Musk rhetoric on this forum?

Clip showing the vehicle was indeed without a driver in the seat: https://twitter.com/EHarringtonTV/status/1659780799716261894...

So technically and officially bots are leading humans now!

Autonomous vehicles should be required to have a “kill zone”: a little square—away from any potential passengers—that cops can shoot to disable the vehicle entirely. If it were a human driving into a crime scene they wouldn’t hesitate, how come when it’s a driverless car all they can say is “oh, hey, stop… what should we do here?”

This is a terrible terrible idea.

Sure it’s extreme, but I’d argue the “terrible terrible” idea is allowing the tyranny of beta software to run roughshod on the road, while the humans have to cope

This is the most American solution to a problem that I have encountered for quite some time.

Someone doesn’t know how guns work, and it’s not the wombat.

Not only an American solution, but also an American problem.

It's called the tyres.

Reply on news.ycombinator.com