Hacker News Reader

Comment by switz | Hacker News Reader

The automated car has driven over 200,000 miles and has never once been in an accident. I don't know many humans that can live up to that.

View on HN

200k miles is nothing. Over 8 billion vehicle miles are spent per DAY in the US. One person is killed for every 75 million miles driven. 200k miles isn't enough to test every terrain, under every weather condition, in every lighting, etc. There's too many variables in the equation.

The Google car won't be truly safe until it has logged 1000x miles.

Its not that Google's automated cars haven't been in an fatal crash, its that they haven't been in a crash at all. The US _crash rate_ is a bit over 5 per million miles traveled, you you would expect about one crash in the 200k miles already covered if it was a human driver. The circumstances probably aren't identical, but what evidence we have suggests that these automated vehicles are at least comparably safe to a human driver.

If we expect 1 crash per 200k mi, and the car has driven 200k mi, the evidence we have has sample size 1 ...

We only have a sample size of one for the 1 crash per 200k miles case. So the evidence that these cars are actually _better_ than humans is non-existent. However, the evidence against the 1 crash per 100k miles case where the car is no longer comparable to human error rates is at least suggestive. And the evidence against the 1 crash per 20k miles case is actually pretty decent. I chose my words carefully.

We can also look at it from the perspective of the geometric distribution: if the P(Crash) = 1/200k for each mile driven there is around a 36% chance that the car would have made it 200k miles without a crash.

You can't be serious. The evidence may support it, but the evidence is incredibly limited.

Can you imagine the potential liability Google faces if they try to bring this to market without having done due diligence? This is a car, not an iPad.

It really depends on the implementation. You're assuming that all of the functionality has been revealed. For example, it's not completely out of the question that their technology will b, in part, completely mapped and trained co-ordinates (hey, they did it with StreetView). Each road, each corner, each nuance all recorded several times over for redundant accuracy - plus all the sensory, awareness that they have previously shown. There's also the point to be made that the car may know when and where it will have problems (calculating multiple steps ahead) and let you know that a known obstacle is coming up and you'll have to take over. Nothing wrong with that.

No one said the car should ship tomorrow, but the reasoning: "Because you don't ship when it still kills people" is ridiculous. Cars have killed people for 90 years and they still ship. And not every death is due to human error. The fact is, the Google car has yet to even be hit by another car, let alone hitting something itself. It hasn't even remotely injured anything, let alone killed a human being.

The Google car has been hit by another car and it's not ridiculous at all to not ship something that can very easily kill someone (and in this case, almost assuredly innocent bystanders) before you've had time to work out the bugs. I'm glad car companies don't release their drive by wire systems out for public betas too.

http://www.technolog.msnbc.msn.com/technology/technolog/goog...

It's a silly debate anyways considering all the extra hardware you'd need.

> The historic moment passed quickly however, along with any opportunity for robot rebellion hysteria, when Google stated that the car was not in auto-mode at the time of the fender bender.

The Google car is only driven in controlled conditions in sunny California. For example the car simply doesn't work when it snows, so they just turn it off for that duration.

The project looks promising but it still has a long way to go.

But remember: a car that can drive for you perfectly well when the weather is fine but that requires you to take the wheel under adverse conditions is still an awesome product as long as the limitations are clear, consistent, and understood by purchasers.

I wonder how well we will handle the adverse conditions when we have been letting the car do all the work in fair weather scenarios? "Google wants me to drive in this weather? I haven't driven for 2 years! Now let me see - where is that indicator stalk?"

Where have you found fair weather for 2 years?

It's also a product for which there's a considerable sampling bias in making safety assumptions.

The more so as it's adversely selecting hazardous conditions for the human driver.

My own stats, space travel excepted, are about a quarter million miles driven. Three parking scrapes (insufficient clearance / hitting unseen object), twice rear-ended (neither major damage) both times by a trailing car driving too close, too fast.

Touch wood.

http://techcrunch.com/2010/10/09/google-automated-cars/

Says that they do have the occasional human intervention. So even if it has never been in an accident up to that point there had only been 7 cars that had driven 1,000 miles without any human intervention.

You'd have to have a lot more data (such as a much higher number of miles logged without human intervention and 'all weather' exposure) before you could make such a grand claim.

In very carefully controlled conditions with carefully controlled equipment. Supervised by experts. That's different than a bunch of people modding their Civics and having them roam the streets.

It has been in accident. It wasn't in one it caused but it was rear ended.

Reply on news.ycombinator.com