Uber vehicle reportedly saw but ignored woman it struck • Hacker News Board

Re: Uber vehicle reportedly saw but ignored woman it struck

#21

Post by syncsynchalt » Mon, May 07, 2018, 6:34 PM UTC

retSava wrote: post #5

tlb wrote: post #4

The disturbing thing about this crash is that it's such a basic thing to test on a test track. A pedestrian walking in front at night, with nothing else unusual happening, should be on the list of scenarios to get 5 9s reliability on before taking it out in public. The things you have to test in public are the complex real-world interactions, like these: https://medium.com/kylevogt/why-testing-self-driving-cars-in...

That, and the inattentive safety driver. It's quite infuriating really, letting lose a ton and a half of fast-moving mass without even the decency of looking where it's going. There's really no excuse.

I both can and can't blame the safety driver.

To expect full attention from a person that is not supposed to have anything to do for hours on end (different from driving, which requires constant inputs and adjustments) is not something that's compatible with humans.

With that said they were looking at their phone instead of zoning out eyes front.

Re: Uber vehicle reportedly saw but ignored woman it struck

#22

Post by GenerocUsername » Mon, May 07, 2018, 6:35 PM UTC

freditup wrote: post #18

I'm curious what degree of "cultural context" is built into automated cars - I feel that there has to be a lot of locale-specific techniques for driving. For example, in NYC, there's nothing at all uncommon about a pedestrian beginning to cross the street in front of you and approaching the area of your lane where you will shortly be. Pedestrians typically then stop about a foot away from your lane, let you drive pas…

And welcome to why self-driving cars will always be 3 years away.

I think we should be targeting self driving for long-haul freeway scenarios where the rules are relatively simple and predictable. Trying to control for the infinite number of scenarios in cities is a nightmare which will ultimately bog down the development of any company which makes that its goal.

Re: Uber vehicle reportedly saw but ignored woman it struck

#23

Post by w_t_payne » Mon, May 07, 2018, 6:36 PM UTC

I think that I am qualified to say that cost-effective testing is and will continue to be the barrier to self-driving car adoption.

A big part of the solution to this problem is design-for-test; and the adoption and tight integration of a wide range of (quite conventional but sadly often overlooked) testing and systems engineering standards and processes.

Re: Uber vehicle reportedly saw but ignored woman it struck

#24

Post by toast0 » Mon, May 07, 2018, 6:39 PM UTC

> We’re actively cooperating with the NTSB in their investigation. Out of respect for that process and the trust we’ve built with NTSB, we can’t comment on the specifics of the incident.

Hey look, somebody knows how to follow the script! Uber may be grossly negligent, but at least they respect the process unlike Tesla.

Re: Uber vehicle reportedly saw but ignored woman it struck

#25

Post by jstanley » Mon, May 07, 2018, 6:39 PM UTC

ars wrote: post #15

tlb wrote: post #4

The disturbing thing about this crash is that it's such a basic thing to test on a test track. A pedestrian walking in front at night, with nothing else unusual happening, should be on the list of scenarios to get 5 9s reliability on before taking it out in public. The things you have to test in public are the complex real-world interactions, like these: https://medium.com/kylevogt/why-testing-self-driving-cars-in...

> get 5 9s reliability That's it? Humans have greater than 7 9's reliability when driving. (As measured by fatalities per time of driving, and assuming a fatality takes 5 minutes, and average speed is 30mph.) If the cause of a fatality is 10 seconds of failure, not 5 minutes, humans have greater than 9 9s of reliability. Math: 10 seconds / (1.13 / (30 miles/hour / 100 million miles)) Fatality rate is 1.13 per 100 mil…

It's only really 9 9s of reliability if you assume somebody will definitely be killed as soon as you have a 10-second lapse in concentration, which is obviously not true.

In motorway driving you could probably shut your eyes for 10 seconds every 20 seconds and be perfectly fine almost all of the time. Especially at night time, or other quiet times of day.

Re: Uber vehicle reportedly saw but ignored woman it struck

#26

Post by smogcutter » Mon, May 07, 2018, 6:39 PM UTC

ars wrote: post #15

tlb wrote: post #4

The disturbing thing about this crash is that it's such a basic thing to test on a test track. A pedestrian walking in front at night, with nothing else unusual happening, should be on the list of scenarios to get 5 9s reliability on before taking it out in public. The things you have to test in public are the complex real-world interactions, like these: https://medium.com/kylevogt/why-testing-self-driving-cars-in...

> get 5 9s reliability That's it? Humans have greater than 7 9's reliability when driving. (As measured by fatalities per time of driving, and assuming a fatality takes 5 minutes, and average speed is 30mph.) If the cause of a fatality is 10 seconds of failure, not 5 minutes, humans have greater than 9 9s of reliability. Math: 10 seconds / (1.13 / (30 miles/hour / 100 million miles)) Fatality rate is 1.13 per 100 mil…

I agree with the basic point, but does the 1.13 per 100 million miles figure include freeway driving?

Re: Uber vehicle reportedly saw but ignored woman it struck

#27

Post by carlosdp » Mon, May 07, 2018, 6:39 PM UTC

leggomylibro wrote: post #13

_ea1k wrote: post #3

Earlier quoted context omitted.

I think so. But I'm not sure why that would be a bad thing. The Uber built system was designed to fully replace such systems.

As long as Uber's system logged which automatic brake was triggered - or simply failed to log that its own braking system stepped in - it could have saved a life without impacting the data gathered. They didn't put a tarp over the windshield; why get rid of other proven safety features?

You don't generally want multiple autonomous control systems in a single vehicle competing with each other for control. There are scenarios where the less-sophisticated collision avoidance radar could do the unsafe thing where the self-driving system has a more holistic plan.

Basically, there's no reason the self-driving tech shouldn't have been way more effective than the collision avoidance radar other than Uber royally screwed up. The LIDAR and other sensors should have (and by this article, likely did) detected the woman long before the vehicle was in striking distance, much longer range than the forward radar.

Re: Uber vehicle reportedly saw but ignored woman it struck

#28

Post by joncrane » Mon, May 07, 2018, 6:41 PM UTC

GenerocUsername wrote: post #22

freditup wrote: post #18

I'm curious what degree of "cultural context" is built into automated cars - I feel that there has to be a lot of locale-specific techniques for driving. For example, in NYC, there's nothing at all uncommon about a pedestrian beginning to cross the street in front of you and approaching the area of your lane where you will shortly be. Pedestrians typically then stop about a foot away from your lane, let you drive pas…

And welcome to why self-driving cars will always be 3 years away. I think we should be targeting self driving for long-haul freeway scenarios where the rules are relatively simple and predictable. Trying to control for the infinite number of scenarios in cities is a nightmare which will ultimately bog down the development of any company which makes that its goal.

> And welcome to why self-driving cars will always be 3 years away.

It's also possible that human/pedestrian behavior will adapt to the consistent behavior of driverless cars.

Re: Uber vehicle reportedly saw but ignored woman it struck

#29

Post by ShabbosGoy » Mon, May 07, 2018, 6:42 PM UTC

freditup wrote: post #18

I'm curious what degree of "cultural context" is built into automated cars - I feel that there has to be a lot of locale-specific techniques for driving. For example, in NYC, there's nothing at all uncommon about a pedestrian beginning to cross the street in front of you and approaching the area of your lane where you will shortly be. Pedestrians typically then stop about a foot away from your lane, let you drive pas…

> But it seems like a very non-trivial problem to do this correctly.

I had this same thought. You could theoretically have a universal rule where the self-driving car stops 100% of the time if it detects an object that is determined to be a pedestrian. But there are unforeseen problems with this approach. What if the model is overly cautious and there are many false positives? That could be extremely dangerous too, if a self-driving car is constantly stopping and going.

Re: Uber vehicle reportedly saw but ignored woman it struck

#30

Post by assblaster » Mon, May 07, 2018, 6:45 PM UTC

lhorie wrote: post #17

assblaster wrote: post #7

The question I want to know is: Did Uber doctor the camera video to make the road seem darker than it actually was? If you recall, there was initially a doubt that Uber was at fault because the streets appeared extremely dark, but I think they were buying time to delay the outage at incompetent Uber. The police initially didn't fault Uber based on this video.

I highly doubt it, considering the video came from Tempe police, not Uber. Regardless, does it matter at this point? Uber is basically saying here that the sensors did detect the pedestrian but the issue was that the software to ignore false positives was tuned incorrectly.

Tempe police obtained the video from Uber, Uber could have easily altered it before handing it over. Uber has behaved badly in the past, and they had the motivation and past experience to do what it takes to advance their agenda.

Remember, Uber terminated its autonomous driving testing as a result of this preventable tragedy.