Hacker News Reader

Comment by ec109685 | Hacker News Reader

It's not like humans are perfect in parkings lots though.

Humans aren't perfect but the rate of incident for human drivers is surprisingly low, it is well below 1% of all miles driven. Tesla wouldn't even remotely come close to achieving human level safety.

What do you base that on?

Basic statistics

http://www.driverless-future.com/?page_id=983

How in the world does that article say this: Tesla wouldn't even remotely come close to achieving human level safety

A million Teslas on the road generate 10 billion miles a year.

Did you even bother reading the article? Because the point is, as of right now, the number of human driven miles completely and utterly dwarfs the number of autonomous miles. That's why in order to compare the relative safeties, you'd need several orders of magnitudes of more sample data from the autonomous driving side.

Please try to understand the underlying statistics, it's really not hard.

That isn’t what the article says. It concludes with this:

> In summary, this article shows that the idea that self-driving cars need to drive hundreds of millions of miles before we can be convinced that they are safe is full of flaws. It is misleading to just focus on fatality rates where many other correlated measures for reliability are available that are easier to measure. It is wrong to focus primarily on accidents; the focus should rather be placed on the avoidance of safety-critical situations.

Read the first half of the page, it is what it says. It is countering claims like this one:

> For those driving without Autopilot, we registered one accident or crash-like event for every 1.92 million miles driven. By comparison, the National Highway Traffic Safety Administration’s (NHTSA) most recent data shows that in the United States, there is an automobile crash every 492,000 miles.

https://www.tesla.com/blog/q3-2018-vehicle-safety-report

I posted the article to show how Tesla's way of measuring safety/accuracy is flawed. The article does a very good job demonstrating that. I agree with the article that we should come up with a procedure/policy driven approach to validating the safety of autonomous vehicles, because there would be no easy way to validate the safety of autonomous vehicles given the volume of human miles driven.

MobilEye's RSS is aiming to do that.

I bet parking lot incidents are underreported. A minor bump of someone else's car is far less likely to get reported if they aren't around to see who did it.

Hopefully far, far below 1%. If I was hitting someone once every hundred miles, the front of my car would be pretty messed up.

>For those driving without Autopilot, we registered one accident or crash-like event for every 1.92 million miles driven. By comparison, the National Highway Traffic Safety Administration’s (NHTSA) most recent data shows that in the United States, there is an automobile crash every 492,000 miles.

Source: https://www.tesla.com/blog/q3-2018-vehicle-safety-report

Read this carefully: http://www.driverless-future.com/?page_id=983

Any naive statistical argument comparing human performance to autonomous cars will fall flat on its face due to the small sample size of autonomous driven miles in relation to human driven miles.

Reply on news.ycombinator.com