Slashdot Mirror
Nvidia Suspends Self-Driving Car Tests in Wake of Uber Crash (theverge.com)
Nvidia said on Tuesday it will suspend its autonomous vehicle testing on public roads in the aftermath of Uber's fatal crash in Arizona. Uber is a customer of Nvidia's, using the chipmaker's computing platform in its fleet of self-driving cars. From a report: Nvidia had been testing its self-driving cars in New Jersey, California, Japan, and Germany. The company is hosting its annual GPU Technology Conference in San Jose this week, where it is expected to make several announcements regarding its automotive products. "Ultimately AVs will be far safer than human drivers, so this important work needs to continue," a Nvidia spokesperson said in an email. "We are temporarily suspending the testing of our self-driving cars on public roads to learn from the Uber incident. Our global fleet of manually driven data collection vehicles continue to operate."
Why is every tech company thinking they have the domain expertise to get into the car industry?
Tesla is proving they have no idea how to scale manufacturing. This seems like the kind of things you partner with an actual car maker instead of just grafting this on later.
Because at this rate we're going to end up with dozens of different self-driving cars, all of which have their own quirks and warts.
What could possibly go wrong?
Can't patent math.
Someone should get a patent on getting patents.
I think the weaknesses in the driving system are probably more than likely Uber's fault but Nvidia is probably suspending testing just in case there's an issue with their hardware. It's the responsible thing to do when lives are on the line. After all no one wants the negative publicity associated with accidentally killing someone in testing. Note how google's cars haven't run anyone over and at worst have been involved in minor fender dings that still made news.
1. The pedestrian was J-Walking
2. The driver was paying no attention to the road.
3. The sensor wasn't able to respond in time.
I think the nVidia chip is the last thing that should be faulted here. We have two clear cases of human stupidity to blame before the chip comes in. They should just re-brand it as "Computer Assisted Driving" instead of "Self-Driving".
I doubt it has an atmosphere, but it doesn't really matter.
What would nuclear materials matter? I'm guessing the answer is no, given this was a manned station and nuclear materials are kind of dangerous...
I'm guessing the only possible issue is propellant tanks, which are likely empty but could survive re-entry.
Question.. What's this have to do with NVidia and self driving cars?
"File to fit, pound to insert, paint to match" - Aircraft Maintenance 101
At this point I'm just waiting on Nestlé, Pfizer and Coca-Cola to also announce they're pausing their autonomous vehicle development. Is Burger King also working on AV tech?
Your upper middle class soccer mom causes far few accidents than your unlicensed drug dealer bum.
I get licensed vs unlicensed; it seems intuitive that unlicensed drivers would be more accident prone. But the rest? Does being upper middle class improve your driving ability? Does being a soccer mom? Does dealing drugs make you worse? Should we ban pharmacists from the road?
...soccer moms will actually have MORE accidents.
What makes you think that soccer moms are going to outperform autonomous cars? Or do you just mean more accidents relative to how many the "bums" are having?
He's getting rather old, but he's a good mouse.
If you disagree don't bother debating with me, I'm not listening.
Ah, the closed minded approach.
For those on /. who ARE willing to listen, the point is not for self-driving cars to be perfect. Real-world scenarios suggest that a zero-crash world is virtually impossible.
What we need are for self-driving vehicles to be far safer than human-driven vehicles, not necessarily zero crashes. And so far, that has proven to be the case. Right now there are about 32,000 fatalities in car crashes across the US every year. Even if that number were reduced to 20,000 fatalities it would be well worth it in lives saved; it is quite likely if all cars were replaced the fatalities could be reduced from thirty thousand to a few hundred per year. Some would still die but it would be a tiny fraction of those who die under human control.
In this crash the company and police released videos and telemetry. The video shows a jaywalker at night walking their bike (which has no reflectors or lights) across an unlit road after coming out from behind a shrub and small metal fence (about the same density and radar reflectivity as a bike) in the median. The computer's radar identified the moving obstruction and slammed on the brakes about 200 milliseconds before I could see anything in the unlit patch. No human could have seen the pedestrian in time to apply the brakes. If they did notice, many people in high speed collisions will swerve and roll their vehicle as well. The computer managed to apply the brakes, but also realized it could not avoid the pedestrian by swerving, nor could it stop the vehicle from highway speeds in that distance despite slamming on the brakes.
Even though the jaywalker died, the computer still reacted faster and more accurately than any human could. The computer identified and was reacting (including realizing swerving would not help) before there was anything visible to the human eye.
Tragic as her death was, it was caused by physics of speed and the jaywalker's choice to cross in an unlit area, walking from behind a shrub, directly in front of a high-velocity car that was completely visible if they looked to their right. Neither a human driver nor computer driver could compensate for the jaywalker's fatal choice.
//TODO: Think of witty sig statement
Perhaps stricter driving tests would do more good today than attempting to push the envelope on public roads.
They can test all the self driving cars they want in New Jersey, California, Japan, and Germany. I don't live in any of those places.
The vast majority of accidents can be prevented by reforms in driver education, driver training, and driver testing, which includes more frequent re-testing of drivers, to catch the development of bad habits (requiring remedial education and training), and removing chronically incompetent drivers from behind the wheel permanently. For decades now driver education/training/testing has been chopped down to the bare minimum and the effects are showing. Add to that the influx of new citizens from other countries, who weren't good drivers to start with, and it's that much worse. Also add to that 'safety hawks' cherry-picking their statistics to fit the narrative that humans are incapable of driving safely, plus greedy self-driving car companies wanting to rush their 'product' to market as fast as possible, and you get what we had happen here: DEATH. SDCs need to be removed from public roads, and there needs to be reforms in driver education/training/testing. Situations will dramatically improve after that.
Clearly you have not read anything else other than the article from the day it happened. There are so many things we now know about the incident that so directly go against what you are saying I'm not even going to go into them.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
How many people have to die before we start to observe common sense and decide, once and for all, that putting self-driving cars on the same roads with non-self-driving cars is a bad idea?
50,000
Every time we go here, someone observes that "well, we just have to make the roads more hospitable to self-driving cars; we need built in signaling, dedicated paths for them, reflective whatever on all other objects on the road..."
Nonsense. You just need to stop walking out in front of moving vehicles.
Don't tell me let me guess, you are paid by an auto drive company. First and foremost, as soon as you said it was dark you lost me. The car had LIDAR, you know, that thing with it's own emitters that can see in the dark. In fact as I pointed out before, can see better in the dark because the sensor catching the return has better SNR without the sun. Secondly, why do these cars not have infrared detectors? I can get one on a BMW I know and I image many other premium cars. Or is uber too cheap? You know, like they were too cheap to have two people in the car like they used to. It was murder, someone needs to pay.
And lastly, the whole accident stat thing is a nonstarter at this point. There are not enough numbers to make it work. Depending on how you "work" the numbers, I've seen analysis where self drive are more dangerous by a factor of 3. As has been said often, statistics don't lie, liars use statistics.
26. The number of people killed by non-self-driving cars in 1899 (the first year the USA kept records on such things). IOW, about three orders of magnitude lower than were killed by non-self-driving cars last year.
Note that if we'd used the same sort of "common sense" in 1899, we'd be using horse and buggy today....
"I do not agree with what you say, but I will defend to the death your right to say it"
What ever happened to Mankind being able to do great things and solve hard problems
We did it and we continue to do it by stepping on the brakes and analysing the situation before continuing.
Do you know how many people were lost during the building of the Panama Canal
Do you know why we are able to build such projects with far less loss of life now?
We've lost our balls and our teeth.
You're an idiot if you think that.
If you disagree don't bother debating with me, I'm not listening.
I stand corrected. You're not an idiot at all. You're a fucking moron and a danger to everyone around you. You have no concept of how the world works or how it worked. Keep living away in your bubble of ignorance.
What is parent +5 insightful? Its full of false assumptions and outright lies.
I find it kind of odd seeing how the Uber car had both LiDAR and sonar sensors that should have detected the pedestrian despite the darkness, but the car still didn't stop. The obvious suggestion as to why it didn't stop was that the control system simply couldn't react fast enough to the input from these sensor inputs, but I have a feeling it may be something else.
The thing I don't like about Tesla's Autopilot system is that at even slightly higher speeds it relies completely on the visible light cameras and image recognition to identify obstacles, other motorists, road signs and the contours of the road. This has lead to at least one fatality when a Model S driven by Autopilot crashed right into the back of a tractor trailer for the simple reason that it just couldn't tell the metallic grey color of the trailer from the sky. The company behind the tech Tesla's Autopilot system is based on, Israeli company Mobileye, had been telling Tesla for quite some time that they were overselling their system and using their tech for something it was never intended for and this was the final straw for them.
The way Tesla ties into this fatality is that Tesla chose Nvidia's Drive PX control and compute units for self-driving cars as their replacement. Like the Mobileye tech before it, the Drive PX-based system also builds first and foremost on image-recognition with visible light cameras and it's exactly the same tech that's used by Uber in their self-driving car program after they had to dump the tech stolen from Google. Thus to any system based on this tech, additional LiDAR and sonar sensors are just tacked on and thus not going to be acted upon as quickly as data coming in from the visible light cameras. I wouldn't be the least bit surprised if Uber hasn't even had the time to properly implement support for the LiDAR and radar sensors on their test cars into Nvidia's system.
Google/Waymo's self-driving tech on the other hand is primarily built around the LiDAR system, meaning that decisions made based on data from it is going to be acted on very rapidly. Unlike visible light cameras it's an expensive sensor, but it's not hampered by lighting conditions.
"Why should I want to make anything up? Life's bad enough as it is without wanting to invent any more of it."
Numerous people have gone to the crash site and reconstructed it. There was plenty of light for a human to have seen the victim and to have stopped in time.
When a person makes a mistake they learn. When a mistake like this happens all autonomous cars learn. This needs to press forward.
They can press forward all they want, just not on public roads until they can demonstrate the cars don't suffer from night darkness or any other kind of flaw that makes them worse than a human.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
er, night blindness*
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Humans drove 3.22 trillion miles in 2017 in the US and there are 32,000 deaths a year. This means all self driving cars need to drive 100,625,000 miles without a death to be as safe as a human. I wonder how close the Uber car got?
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Private companies should get out of technology that enhances our lives. It should be reserved to car companies who turn wrenches and government agencies. What can go wrong?!?!ðY¦ââ(TM)ï
Or we'd have someone walking out ahead of the self-driving cars carrying a flag.
Check out my sci-fi/humor trilogy at PatriotsBooks.
Even though the jaywalker died, the computer still reacted faster and more accurately than any human could. The computer identified and was reacting (including realizing swerving would not help) before there was anything visible to the human eye.
This is simply a lie. Any half witted human driver had at least 6 seconds to notice the pedestrian. Sadly the only person in the loop chose to look away from the road for most of those 6 seconds, and by the time they looked up it was too late.
Besides, Uber can't even manage its own target of THIRTEEN miles between interventions. Those are not self driving cars in any way.
This was manslaughter, and if Uber wanted to destroy the future of self driving cars, they could not have done a better job of it.
Finally! A year of moderation! Ready for 2019?
No one can learn from a self driving car that only manages 13 miles between interventions. It should not be allowed on public roads in that condition; with such a short time between interventions there will be lots to learn from driving on closed tracks.
Once a car can do, say, a thousand miles between interventions on a test track with simulated obstacles, you can let it out where it can kill people.
Dara Khosrowshahi needs to go behind bars for manslaughter.
Finally! A year of moderation! Ready for 2019?
and removing chronically incompetent drivers from behind the wheel permanently.
But my freedoms! You suggest we travel down the path to a totalitarian state where I'm denied my God given right to drive just because I get into a few accidents each year.
“Common sense is not so common.” — Voltaire
Dang. I was just going to patent the corollary to the Pythagoras Theorem for calculating Angle-Side-Angle distances from a point projected out in the middle of the triangle. Then no autonomous car could calculate distances and they would all crash.
Note that if we'd used the same sort of "common sense" in 1899, we'd be using horse and buggy today....
Also note that it is almost certain that more than 26 people were killed in the US by accidents involving horses in 1899.
Finally! A year of moderation! Ready for 2019?
46. That was life expectancy back then. You're saying you want to go back to those days?
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
What the fuck kind of nonsense is that?
Either it needs to continue, or it needs to be suspended.
Or are they saying that the work exists in some kind of superposition of both states simultaneously?
File under 'M' for 'Manic ranting'
You suggest we travel down the path to a totalitarian state where I'm denied my God given right to drive just because I get into a few accidents each year.
Yep. Enjoy your Uber, or Lyft, or cab ride, or public transit. No reason the rest of us should have to endure imminent death at the hands of shitty pseudo-intelligent machines just because you're too stupid to learn to drive safely.
Any half witted human driver had at least 6 seconds to notice the pedestrian.
Six seconds earlier, as seen from the video?
Half-witted drivers must have learned to see in the dark.
When I watch it, the first moment I can see the pedestrian is the light shoes at the very end of 0:06 into the clip. In the one they released with the computer's response (my Google search is failing me for finding it) the computer slammed on brakes about a quarter second before that moment, when the pedestrian was just about to enter the lane.
//TODO: Think of witty sig statement
I'm just going to leave this here and say, that a really likely outcome of this investigation is that Uber is a shit company. Just my two cents though.
It's people doing 55 when others are doing 70-75+ that leads to crashes
Who said anything about math?
Nvidia already has a decent amount of experience with physics calculations through PhysX. They have experience with creating boards that can do a LOT of computations very quickly in their graphics cards - enough so that it's how stuff like Bitcoin got started.
Now they're putting some cars on the road to see what works and what doesn't. At a guess, they want to create a system that can quickly calculate a dozen or so steering solutions in case of an emergency, then choosing the one with the least collateral damage based on physics calculations. Why just hit the brakes if you can win another couple of feet of minimized braking distance by applying more braking force to one side and turning the wheel, effectively causing the car to come to a skidding sideways halt?
-=This sig has nothing to do with my comment. Move along now=-
The video is manipulated or the camera is useless. It is obviously possible that it is footage from the actual camera that Uber uses for its computer vision -- which would only make it even easier to convict Uber's CEO of manslaughter. However, any half decent video camera would have spotted that pedestrian, and humans are much better than cameras at low light. There are other videos showing how well lit the road actually is.
Finally! A year of moderation! Ready for 2019?