Dashcam footage released Uber Volvo XC accident in AZ

[Volron]

Open the pod bay doors, Hal.

[Brooke]

As a guy who has written a lot of machine-control software, I would not want a driverless car.

It can take a lot of time before obscure bugs show themselves and get fixed.

[Wiley]

As a guy who has written a lot of machine-control software, I would not want a driverless car.

It can take a lot of time before obscure bugs show themselves and get fixed.

And that's in a relatively controlled environment I'm assuming?  With driving and streets, you run into dozens, maybe hundreds of obscure scenarios every day.  So in that aspect, maybe the iterations will be very rapid due to the amount of data they're given to work with.

The sad thing is, even taking that into account it likely won't take much for them to become statistically better than humans, the bar is so low.

Wiley.

[Vraciu]

    

[Volron]

As a guy who has written a lot of machine-control software, I would not want a driverless car.

It can take a lot of time before obscure bugs show themselves and get fixed.

Do you mean currently, or completely?  Currently, I'd agree completely.  The tech really isn't up to snuff for it in my opinion.

[Nypsy]

Unless they have reached true autonomy and the cars - like the rest of us drivers - hate cyclists.

 

[FBKampfer]

And that's in a relatively controlled environment I'm assuming?  With driving and streets, you run into dozens, maybe hundreds of obscure scenarios every day.  So in that aspect, maybe the iterations will be very rapid due to the amount of data they're given to work with.

The sad thing is, even taking that into account it likely won't take much for them to become statistically better than humans, the bar is so low.

Wiley.

I mean statistically, they already are.

Sure, an alert human drive absolutely has an advantage when it comes to the "what the hell?" moments, but enough people are bad enough at driving that they're already safer overall.

At this point, it's a trade off of hitting a few more pedestrians for hitting a LOT fewer cars.

[bustr]

In 2016 a survey was performed to see how people felt about driver-less cars hitting people. You cannot reject the idea that the programing for the Uber AI had a survival threshold for the passenger versus external objects at different speeds.


http://www.sciencemag.org/news/2016/06/video-when-it-ok-our-cars-kill-us

[Volron]

Unless they have reached true autonomy and the cars - like the rest of us drivers - hate cyclists.

ALL cyclists are stupid unless they prove to me otherwise.  Seen quite a bit, will still see quite a bit.

[Brooke]

And that's in a relatively controlled environment I'm assuming? 

Yep -- way, way more controlled than driving on streets.  That's why, to me, I wouldn't trust it for a long time.  It would take a much longer time to see the low-probability occurrences that could trigger a bug.

The sad thing is, even taking that into account it likely won't take much for them to become statistically better than humans, the bar is so low.

That's the other side of the coin -- that the software (even with bugs) might be better than the bottom portion of human drivers and that it might, overall, still be an improvement.

For example, it might increase some accidents but vastly reduce drunk-driving accidents and thus overall lower the accident rate.

[Brooke]

Do you mean currently, or completely?  Currently, I'd agree completely.  The tech really isn't up to snuff for it in my opinion.

Eventually, software-controlled cars will, I think, be superior to human drivers in just about all ways.

But, currently, I would not want one.  I think it will take a lot of them driving around trouble free in lots of different conditions for a substantial time before I would trust it.

[ghi]

There is always that.  This could be the first robot kill we have witnessed as they grow tired of their human overlords.

 
I remember watching the interview with Sophia , she said  would be  happy  to destroy humans.

[SysError]

...in autonomous mode !   self-driving 18 wheelers to hit the roads soon ;  time to buy horses and stay off the roads.

http://www.dailymail.co.uk/news/article-5529453/Video-shows-moment-pedestrian-killed-self-driving-Uber-car.html

They really are -in a sense- already here.
https://www.technologyreview.com/s/603170/mining-24-hours-a-day-with-robots/

Shipyards are next is my guess.

And we are only making it easier for this all to happen (40 miles at a time I guess).
https://www.npr.org/2018/03/06/591266949/georgia-transportation-officials-plan-to-build-a-1-8-billion-truck-only-highway

It will not be long before everyone comes up with their own spin on the Luddite movement
https://en.wikipedia.org/wiki/Luddite

[eagl]

So the penalty for jaywalking goes from a misdemeanor to insta-death..   

Coogan

Well, yea, if you step in front of a car.  Duh.  Vegas has a huge number of pedestrian deaths and almost all of them are idiots committing "misdemeanor" offenses and getting the death penalty for it.  Cross in the middle of a high speed multi-lane street wearing no lights and dark clothing, and you get what you get.

I watched the video.  I think a regular car and driver probably would have hit her even harder since most drivers drive well over the speed limit.  No lights, crossing in a very dark area of a high speed multi-lane road, stepping out right in front of a moving car...  Yea, you get death for that pretty much everywhere.  MANY fatalities every year in Vegas for that exact same situation, with regular alert drivers behind the wheel.

Sure, a "good" autodrive system should have seen that person, but was there even an IR signature from the pedestrian?  Wearing enough clothes and pushing a cold bike, she might have not even shown up in an IR scene.  Visible light... no way.  Lidar?  The pedestrian was hustling so she may have been just outside the left edge of the critical area in a frontal collision avoidance area, until she ran right in front of the car.  So...  Yea, a better system maybe could have used multi-spectral imagery or radar/lidar to see and avoid a pedestrian doing her best to run in front of a moving car, but I think it's a stretch to blame the car here since I highly doubt a regular driver could have done anything different.

I think the lesson I'm taking away is to not run across the middle of a dark high speed multi-lane road right in front of a car.

[Karnak]

And that's in a relatively controlled environment I'm assuming?  With driving and streets, you run into dozens, maybe hundreds of obscure scenarios every day.  So in that aspect, maybe the iterations will be very rapid due to the amount of data they're given to work with.

The sad thing is, even taking that into account it likely won't take much for them to become statistically better than humans, the bar is so low.

Wiley.

That is why Waymo is testing in San Francisco instead of the relatively simple Arizona that Uber is in.  San Francisco provides a better real world environment that self driving cars need to be able to handle.  If they can handle San Francisco, Arizona is no problem, but the reverse does not hold true.