Self-driving car accidents revealed in California | Hacker News Reader

Self-driving car accidents revealed in California

bbc.co.uk

15 pointshaack21 comments

Well that's a non-story.

And it's not even consistent with itself.

The Delphi car was stationary when hit, it says, and then later says they were all "going very slowly" during the collisions.

this seems a little link-baity to me.

Given most of the accidents were fender benders CAUSED by human drivers it seems to support the thesis behind self-driving cars...

Well, spokesmen for the companies operating the driverless cars claim that they were not at fault. As far as I can tell from the article, there is no independent confirmation of that assertion.

Smooth traffic happens when all drivers are doing it well. This is part of why self-driving cars could work... they will work better together than humans do.

But if humans are hitting them, what we need to know is why. See, my stream of consciousness on this goes something like:

Was the human just being careless? Or did the car do something that would be safe under normal circumstances, but didn't fit the actual scenario, so a human was surprised by their actions? For example, if a piece of debris flies in front of a self-driving car on the interstate, does it just stop? Or does it check behind itself, and the lanes next to it, to see if someone behind it is tailgating, so moving to another lane is safer than stopping? Do they activate their hazard lights when something is wrong up ahead, to alert the tailgater behind them? Can they judge lateral motion of an object well enough to make the call on whether the flying debris would even still be in their lane when they pass? If they judge it will miss them, but hit the guy next to them, do they move over to avoid the carnage? If they do make a lane-changing move, do they do the move I see pickups do all the time, swerving out of the way of a stopped object/car, so the person behind them is suddenly presented with the stopped object, but with way less reaction time to avoid it, thereby making the human-powered car have an accident just because the self-driving car, frankly, is driving like a dick?

I have no idea how deep the logic goes in accident prevention in these cars. But there is a lot more to it than just not hitting something. I'd love to see an in-depth article covering that aspect of self-driving algorithms.

In case you want some other sources of the same "story":

https://news.ycombinator.com/item?id=9524678 (nbcnews.com)

https://news.ycombinator.com/item?id=9524737 (startribune.com)

https://news.ycombinator.com/item?id=9524859 (ap.org)

https://news.ycombinator.com/item?id=9525344 (sfgate.com)

https://news.ycombinator.com/item?id=9525852 (seattletimes.com)

https://news.ycombinator.com/item?id=9526220 (yahoo.com)

It's pretty clear that the self-driving cars aren't at fault. They even quote an anonymous source that says the cars were occupied and were moving at slow speeds. Nothing to see here. Move along.

The story here appears to be that none of the cars were at fault and were stationary when other cars hit them - not really an indictment of autonomous vehicles as the title suggests.

In the last 8 months: - Number of human-driving auto accidents in US: ~10.6 mil - Number of self-driving auto accidents in US: 4 Guess which one merited an article? --- Editing to add that all 4 crashes were apparently caused by human drivers https://twitter.com/msquinn/status/597738756511277056

There are 255m cars in the US (1). 10.6m accidents represents 0.04 accidents per car.

4 out of 48 self driving cars in California have been involved in accidents which is 0.8 accidents per car.

(1) https://en.wikipedia.org/wiki/Passenger_vehicles_in_the_Unit...

I think you also need to consider the number of regular cars vs the number of self driving cars.

Yeah, rates are important, but also small sample size (and who's at fault matters - the companies claim that human drivers caused all 4).

So you're not wrong, but my point is that humans are awful drivers, prone to distraction and fatal crashes. (32,719 in 2013 alone). If anything else (terrorists, sharks, aliens) were causing 30k deaths a year, we'd wouldn't be blithely accepting it as the status quo and acting fearful of measures that could dramatically ameliorate the problem.

Tens of thousands of people die in car accidents due to human error every year[1], but society won't accept a single death from a self-driving car[2]. Interesting dynamic, isn't it?

[1] http://en.m.wikipedia.org/wiki/List_of_motor_vehicle_deaths_...

[2] This article.

> society won't accept a single death from a self-driving car

What has convinced you of this?

Because an accident caused by a self-driving car will have huge financial consequences (liability) for the manufacturer. So there's no way self-driving cars will ship when they merely reduce the total number of deaths. They'll need to be nearly perfect.

Can you imagine the headlines if there were thousands of deaths caused by self-driving cars even if it were say a 10x reduction in total deaths?

400 out of the 4800 driverless cars were involved in accidents last month. None of them were the driverless cars' fault :-P

Just kidding. I'm glad no one got hurt!

What is the comparison to human-operated vehicles?

> "Safety is our highest priority. Since the start of our programme six years ago, we've driven nearly a million miles autonomously, on both freeways and city streets, without causing a single accident," said a spokesperson.

Assuming the spokesperson is correct here, this would seem far better than most humans on the road. According to this article[1] (also about driverless cars):

> To put that into perspective, the average U.S. driver has one accident roughly every 165,000 miles. Here's how we got that figure: our average mileage per year is 16,550, according to the Federal Highway Administration; the average length of time we go between traffic accidents is 10 years, according to Allstate.

And since, according to the article, it doesn't seem like they've yet caused accidents, I would presume that "million miles" figure is still on the rise.

[1]: http://mashable.com/2012/08/07/google-driverless-cars-safer-...

Isn't this mixing two different things? The average driver "has" one accident per 165,000 miles, driverless cars haven't "caused" an accident. The current article discusses accidents not included in this Mashable calculation. Even if driverless cars don't "cause" accidents, they may be involved in more accidents per mile (I don't think we know yet publicly). It may be the case that human drivers are better at avoiding unexpected dangerous situations, for example.

These were human-operated vehicles.

Four out of 48 seems incredibly high considering the 1.5 year timespan. On the other hand, 48 is a small sample and these cars might be driving a lot more miles than the average car

Reply on news.ycombinator.com