Nvidia halts tests of self-driving cars after Uber accident
Click here to post a comment for Nvidia halts tests of self-driving cars after Uber accident on our message forum
fantaskarsef
Hmm unless it's a calculation error I don't really see why this seems to be NV's problem. I guess they want to make sure that their product calculates right, and control the damage done to the future market they want to cater to.
scatman839
In the video the person is just pitch black not visible until the last moment when in the headlights, not sure what more the car could have done. Radar malfunction?
Also the reuters link is broken hilbert
Fox2232
Fatal incident with pedestrian. it was not high speed highway, right? So there was already speed limit in place. And car simply did not detect pedestrian and did not slow down a bit. Otherwise it would be very unlucky to kill at reduced speed.
So for now Pegasus is more like Icarus. (That goes for all self driving cars.)
Ultimately they'll have to deal paintings on back/sides of other cars which may prove to be confusing for AI. Car with color of the road will be hard to detect.
Car having image of person on its back may put AI driven vehicle to full stop and cause accident for vehicle behind.
Painting of some scenery on side of truck may prove to be fatal too.
Those trucks having additional vehicles on board are again quite some danger.
1st thing where failure comes from is insufficient detection of unknown situations. Self learning AI will try to adapt to new situation based on prior memory. But that should automatically be considered as dangerous situation.
Hilbert Hagedoorn
Administrator
Fixed thanks, and yeah I agree. I have seen the video as well. No human could have responded and prevented that. However, you could make the arguement, an autonomous car should have prevented it with all its sensors?
wavetrex
Most likely a human driver would have ran over the pedestrian as well.
There's hundreds or thousands of deadly accidents every week all over the world which aren't making the news anymore, because ... well, we got used to the fact that cars cause fatalities.
But when a self-driving does it, SHOCK, HORROR, a robot car hit some dumbass who crossed a busy street far from the crosswalk in the middle of the night without bothering to care to look if any cars are coming.
Police reports are saying that NO CAR could have avoided that accident, human driven or not.
Because PHYSICS.
Fox2232
Considering all that was said yesterday in nV's stream. I am sure some sensors should have seen person. That car did not even slow down.
And person could have step on breaks. Car like that can stop in 4 meters on such road as it was not wet.
Anyway:
Radar, IR vision, ultrasound distance detection. Those should have detect person as object in front of car.
It looks like car would have crashed to wall if it was there.
Monchis
And I guess no one put a foot in jail.
Yogi
The car should have been able to see that pedestrian. So that means either faulty sensors, bad code or a faulty nvidia driver update
fantaskarsef
Exactly, it is most likely a malfunction on the technical side. I could imagine that the software had some issues there, more so than the hardware itself.
wavetrex
Re-watched the video.
Yes, the car could have, even in complete darkness (IR, Radar, Lidar), superior senses compared to just our feeble eyes. Something to fix in the future.
However, I'm still bothered by the fact that the poor woman didn't do ANYTHING to avoid her own death. Just kept walking.
So I replayed the scene several times... until this single frame which is "sharp" enough
http://dl.wavetrex.eu/2018/accident.jpg
She WAS LOOKING COMPLETELY THE OTHER WAY.
In the next frames it's visible how she turned her head just 1 second before being run over.
She basically walked forward on a street, in the middle of the night, not looking AT ALL towards what were two very obvious headlights of an approaching car.
~~~
Unfortunately, this is quite common, I've seen in many occasions people crossing the street completely ignoring what's going on ! Eyesight straight forward, or worse, with a smartphone in their hand, zero attention to the outside world.
And because of people like that, someone poor chap goes to jail for manslaughter just for being at the wrong place at the wrong time driving home from a crappy day at work.
gUNN1993
Also remember there WAS a human in the car who didn't react. One would assume their job is to be extra vigilant during this testing phase.
Bojan
We all here are aware that cameras tend to, and in fact most of the time, not capture all the details, even more so in poorly lit environment. The camera footage of an incident had so poor quality, that I would make no assumption to weather a human could or could not see the pedestrian in the same environment setup. But even with that poor quality footage, we actually CAN see the pedestrian from a greater distance of that when it was lit by car lights, especially at few moments pedestrian was breaking line of sight with some street lights in the distance. Now, since we can get a glimpse of that pedestrian trough bad camera lenses from the video, imagine how much more aware one would be if it was actually behind those wheels. The whole darkness argument is useless if assumed only by watching that poor quality video. The main issue, in my opinion, is the person behind the wheel. It was obviously not paying attention to the road, instead watching some gadget, only to realize there is danger when it was too late. I am 100% positive that had that person done what it was supposed to (watch the road), no accident would occur. This was a test drive, and the human presence was required because exactly these situations.
Ssateneth
This is what happens when nvidia volta miscalculates a number.
Noisiv
Seems to me that everyone is at fault here:
1) The poor woman....crossing the road, wrapped up in her thoughts
2) Driver not paying attention
3) Those autonomous visibility/speed setting seemed awfully relaxed, especially considering it's a testing phase.
And on top of that you have 2 entities in charge (the man and the computer), and the computer seems quite confident, going 40mp at the THAT visibility, because hey the human is overseeing the entire operation right?
and the driver is relaxed as well and without a worry in the world because, because hey why should I pay any attention when I have computer doing all the work..
And one final thing...if zero car accidents was our number #1 priority, then no car should be allowed more then... 20mph
In other words, if we are going places at normal speeds - people will die.
allesclar
Snowflake Society 1 - 0 Common Sense Society
yeeeman
I am sorry to say, but at 60kph, an attentive driver would at least turned the wheel as fast as possible to the left and step on the brakes.
We cannot judge visibility only based on footage because our eyes have much better vision on dark conditions compared to that dash cam.
RavenMaster
So much for yesterday's conference.
allesclar
She did not cross the road at a proper location (crossing etc) and did so with no care or attention.
A driver at night should be more assertive owing to the conditions, but typically is less due to tiredness.
What more can you do? What if the vehicle had swerved into incoming traffic and caused a pile up? What if the vehicle had swerved the other way onto a pathway/side walk and taken out innocent walkers?
NewTRUMP Order
https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcSzwNO3FWYXS1fwy-y062Yj3t6NebJTdCtDQJbckKMLJC2suELQ Maybe the computer thought it was seeing a painted icon instead of a person in the roadway?
When I first looked at this freeze frame photo I thought the image was laying on ground flat. Perhaps the computer processed the same thing?
http://dl.wavetrex.eu/2018/accident.jpg
G0dly
Multiple things seem to have gone wrong here and I think the video is a bit of a red herring, there are plenty of dash cam clips available online now showing what that piece of road looks like at night and it is fairly well lit. Also the radars should have easily picked up an object of that size.
https://dgit.com/residents-recreate-crash-uber-car-case-56161/
