Nvidia halts tests of self-driving cars after Uber accident

At least in some countries, jaywalking is illegal. I know, it's not the same thing (nor does it actually help), but sadly, people would just override/ignore it to avoid the notification like they do with seatbelts or safety googles or Clippy.

lol, now I'm imagining Clippy asking that, except paraphrased as "It looks like you're trying to kill yourself. Do you need assistance?"
Ironically, you then kill yourself just to get away from Clippy.

 

How about using a custom Volvo XC90 test vehicle with a sensor array that includes a spinning LIDAR on top, straight from Volvo themselves. But any ordinary Volvo running robust VOLVO software/hardware would have stopped on the spot in that situation as it was really straightforward. This accident really shouldn't have happened.

Spoiler

 

Do you even know what Uber's sensor array uses or are you just a stooge for Volvo? Uber is already using a very similar approach, which includes (but isn't limited to) LIDARs. Volvos can't defy the laws of physics. Just because something is seen, that doesn't mean it can just stop in its tracks. So no, Volvo's sensor array probably wouldn't have made enough of a difference and their cars wouldn't have been able to stop in time.
To clarify, I'm not dissing Volvo here - they're one of my favorite car companies. But even they have had their autonomy systems fail (whether or not deaths occurred, I don't know).

 

He knows because he posted a picture of a UBER Volvo in the spoiler tag. You can see the LIDAR puck and stuff on it.

That being said, I don't know whether Volvo's system would have properly braked or not. He makes it seems like he certain it would but there are plenty of videos online of both Mobileye and AutoLiv's systems failing on various vehicles in pretty obvious accidents. From my own experience, I have a Hyundai with Mobileye's forward collision and I hit a deer ~6 months ago, I managed to manually brake and only ended up clipping it's hind legs at relatively low speed (minimal damage to the car but the deer had to be put down =( )but my collison system didn't kick in at all. So I'm not convinced that this particular incident proves Uber/Nvidia's system more/less prone to failure than anyone else's. You'd have to look at stats from every self driving car vs what's out there now.

 

The inevitable who's to blame for this is going to have some serious knock on effects for self driving cars.

Is the car to blame? It was meant to be in self drive mode aren't they supposed to be safer than a human driver why did it not only fail to not hit the pedestrian but in fact failed to take ANY corrective action? What resulted in this utter failure? The tech, the parameters, did something in the car fail to work or break down, the expense of the tech used and at what point does cost of production vs performance come in to play and IF the cheaper tech is going to be less effective and if it IS the car then who's to blame, the guy who designed it, the guy who built it, the guy who programmed the AI, some guy in purchasing who went for the $50 radar sensor in stead of the $200 one?

What about the driver? Is she to blame it is very clear that she isn't paying attention, that she herself took no corrective action until she hit the pedestrian. She was clearly in a full state of the car is doing the driving so I don't have to pay attention. If she is to blame then who is going to want to buy a car that self drives but requires the driver be fully alert and aware of what is going on at all times so they can take over, seems like an added expense for not much benefit. The other question is could the driver have avoided the accident? In this instance it is hard to say. The camera footage is pretty piss poor so doesn't really indicate what was and wasn't visible to the human eye, I am guessing this is something that will be debated over for a while.

Whatever the result this is going to have an impact on self driving car tech, I am sure that most companies who sell 'self driving' vehicles have disclaimers that say the driver or owner of the vehicle is totally at fault if things go t*ts up. Which is all well and fine if the tech was being used as an addition to the drivers full awareness but as I have said for a long long time the second drivers are told that something is autonomous and automatic and they don't have track or monitor the state of the system themselves they will and do just stop paying attention to it. Ask any number of people with auto lights how to turn the lights on or how to turn on their fog lamps and they will just hit you with the statement 'but they turn on automatically' and what this video her shows is what happens when you take this level of automation to degree of a 'self driving car'. The driver, the person responsible for the vehicle spends almost 8 / 10 of the video looking at anything other than the road ahead, it would be questionable if she was in fact the 'driver' and not just a passenger.

Car makers need to stop automating things and they need to stop selling cars on the back that it will do everything for you, by all means include the safety systems but do it in such a way that the owner is not aware of it. That way you still have the driver being ever vigilant of what is going on and the state of their vehicle, being able to take corrective action if it goes wrong but you also have the hidden safety systems working in the back ground to assist when things go really wrong.

 

Actually, Volvo loves their safety status. Maybe lawsuit for tarnishing brand Name is already in preparation.

 

Even if indeed no car would be able to avoid this collision, the problem is completely different. This car has made absolutely no attempt to avoid this collision, and this means only two possibilities - damaged sensors (I don't think so), or faulty software. This car should not be on the road at all.

 

Obvious none of you realize this, even if she was in a crosswalk, the car would have failed to stop. It failed to stop regardless of any circumstances of the women, regardless if she was in the wrong of where she was crossing.
A human would have attempted majority of the time to try and stop, most of the time still hitting her, but death may have not happened.
Personally I am 100% against autonomous cars. Period. I already hate all of the BS added on to cars here in the US with automatic braking, which has already caused accidents where the vehicle slams the brakes in traffic misjudging. This happened to my friend with a cat running across the road in his 2017 Chevy and was rear ended. Car came to a complete stop and the car was about 30 feet in front of him.

 

Yeah but over millions of miles the autobraking is stopping 40% of front collisions and significantly impacting car fatality rates. Like if I told you with 100% certainty that level 5 self driving reduces your chance of death in the car by 70% but that remainder can happen anywhere, anytime, for any reason and it's completely out of your control.. would you say yes to that? I would but I'm a numbers guy - I think you'll find most people will say no, they'd rather be in control because they think they can do better, but the reality is they probably can't.

 

People MUST be made aware that driving a car is similar to using a table saw. If they don't pay attention while using the table saw, they can get hurt or killed. There are TOO MANY accidents caused by people not paying attention while driving.

 

If that video was real, then yes, Human would still hit her (at reduced speed as you wrote). But video is brutally underexposed. Look at all the light sources and how they do not illuminate anything at all.
Then look at real video from that place. Place is very well lit. Driver looking at road would see crossing person at 300meters.

 

There are already comparative videos, made on potatoes that are MUCH brighter then their video. And if its really true, that they dissabled the collision avoidance systems, then i hope Uber gets pounded withou lube, in the full extent of the law.

youtu.be/CRW0q8i3u6E?t=30s

 

Maybe main control unit in car can't handle two overrides at same time.
There is driver and Volvo's protective override hooked into it. So maybe nV board was plugged into that override slot.

 

Don't know what Uber was using, but the problem with all the assisted driving and smart car's HW is that it has to look nice. So the process of developing a radar antenna is that a designer sends how he want's it to look, then the actual engineer will model and simulate it and will inevitably find that it can never work. So they make changes, which designer has to approve again. Then you have to add on top, that all of these sensors and systems have to be super cheap for mass production. Really, in car industry they negotiate the price in the <0.01$ resolution, as it makes millions in the volumes they make.

So the result is a system that is barely working and is at the physical limit and I am in no way surprised that accidents like these can happen and probably will happen in the future. Maybe we can reduce the car collisions when most cars will be smart and communicate with each other, but detecting people with these cheap ass systems approved by the designers based on how pretty they look is quite an utopian idea.

 

As someone who was rear ended last year by a POS in a hit and run doing 40 while I was stopped, I still dont trust self driving cars. They will have errors just like humans, period. All hardware and software fails at some point. Ive even had brakes fail on me and nearly died because of it.

 

Where did it pop up that they disabled the "collision avoidance systems"? What override? I may have missed some intel on this one.

But if they really disabled any collision avoidance system, isn't it clear that the car doesn't care about anything, just keeping the lane / direct, meeting speed limits, not much more? That would be utterly stupid and quite defeating the intention to test such a car in the first place... if you disable systems like that, you might as well not test such a car at all, especially not with a driver that doesn't even look at the road...

 

I get that but the point is they will have less errors than a human driver. So yeah, while they will fail and it sucks when they do, they will fail less, saving more lives. People just don't like the idea that it's out of their control.

Texter is saying the Volvo cars they are using have their own AutoLiv collision avoidance system that comes with the car. Uber either disables that system and/or utilizes the hardware for its own collision avoidance/self driving. The argument being that maybe Volvo's system would have actually prevented the accident.. but having that enabled kind of defeats the purpose of Uber building/testing their own self driving platform.

 

Ah, thanks for clarifying.

 

Uber settles with family of autonomous collision victim



http://www.foxnews.com/auto/2018/03...-autonomous-collision-victim-report-says.html

 

https://www.anandtech.com/show/12599/nvidia-pauses-vehicle-testing-confirms-drive-wasnt-in-uber?

Figured I'd post this for clarification purposes