Self-operating shuttle bus crashes after Las Vegas launch (All Things Self Driving)

"Uber Halts Autonomous-Car Testing After Fatal Arizona Crash

Uber Technologies Inc. halted autonomous vehicle tests after one of its cars struck and killed a woman in Tempe, Arizona, in what is likely the first pedestrian fatality involving the technology..."

https://www.bloomberg.com/news/arti...nomous-car-involved-in-fatal-crash-in-arizona

 

I want to know what happened too, but accident like this will happen again, as we have no fundamental cure. The cause of this accident will be 'patched', but dev have to 'patch' numerous possibilities and still, one unexpected situation can be game over, well known Frame problem. They halted but will restart as one person's 'unfortunate' death is not comparable for decision maker to billions of $. So how many death will be enough to stop them? Maybe hundreds? The problem is not that they're premature technology, but fundamental inability of these 'AI's for such a complex matter of driving, which shouldn't be confused with e.g. the early days of airplane. But they're very good at assisting human, so 'AI' based accident prevention is not bad.

It seems modern ppl have serious deficiency of understanding death. If you died, that's it and all. But they somehow tend to think "so she died, but I won't" or even "The possibility I'll die by that is 0.2% so it's OK." which is serious misconception of probability too.

 

I am more interested in why the human driver didn't intervene and halt the car.
Mrk

 

This is the kind of situation that cannot be resolved by AI. Even if sensors detect an obstacle 100 times faster than a human, there's still the minimal physical distance needed to stop the vehicle. In machine-human interactions, there'll be always an element of unpredictable. BTW, I even have a patent covering the mixed situation (human and AI drivers and uncertain situations within safe parameters).

Mrk

 

Ofc crossing road apart from crosswalk is bad, as well as carelessly emerge from shadow, but the point should not be who was at fault. It should be if it was unavoidable even if it was careful human driver, which is not yet clear in this case. It may be, but careful driver sometimes can 'see' the risk of such unexpected event w/out actually thinking it. J.Gibson called it affordance. So he might slow down the car and she might not die, but IDK and the driver seated in autopilot mode is not comparable to real driver.

Self-driving car will achieve lower accident rate than human as it is free from human error, but again the point is not that. It is properly securing it is hard. If thinking about what made current security disaster is not enough, there's an interesting example from recent 'AI' Japanese chess match (there're many backstories behind these matches and if you knew, you wouldn't simply think 'AI' won, including that Deepblue). Before the match, a human player was allowed to practice with an AI (it's quite exceptional in the industry) and he found a bug in the AI, so he exploited it in official match and the AI committed suicide! Even noob player don't do such foolish play.

Maybe running self-driving car in airport is OK, as there won't be many unexpected event like public road. But in public road, human supervisor either on behind wheel or remote center, is at least required. But as it will decrease the benefit they'll be finally omitted. Then disasters happen when we least expect them.

 

Tempe Police release interior and exterior view video of fatal accident:

[Warning - Disturbing]

https://twitter.com/TempePolice/status/976585098542833664/video/1

 

I agree with nearly all this reply is suggested as fact. Errors are inevitable in any pursuit. Harm to the human element per moving machinery is always been and shall always realize is present at some point-in-time, which questions where exists a 100% as possible foolproof/failsafe procedure.

In a world increasing if not bordering on complete overload with moving machinery production (Public Motorway Vehicles), distribution, followed by rapid saturation, only IMHO, severely populated municipal towns, especially metropolis cities are the absolute WORSE places to be playing experimental self driving ANYTHING, let alone cars/trucks.

For better or worse accepted opinion, i side with this member those practices might be much better suited for shorter commuting purposes such as airports, perhaps trainways.

 

https://arstechnica.com/cars/2018/0...victim-came-from-the-shadows-dont-believe-it/

 

Note that suicide was only possible because the pro player is far ahead of those devs when it comes to J-chess. Devs simply can't test every possibility even in such a well constrained game. But these 'autonomous car' devs have to expect not only all possible traffic situation, but also all possible attack too and they have to be ahead of any criminal. These cars have to be network connected for various reason, one of them is that precise examination of traffic jam revealed introducing autonomous car don't solve jam by itself but interconnected system of cars and other traffic components is required. And these system are made by human, and...? I wonder how many ppl here can believe such system will be magically secure unlike current IoT (I leave question about privacy to each reader). Another concern is most of 'autonomous' car seem to rely heavily on such a blackbox of NN which also makes it almost impossible to prevent this kind of attacks.

Some online ppl appear to believe 'autonomous' car have better record than human driver as to death rate. But Wired mentioned human driver kill 1.16 per 100M mile while 'autonomous' car killed 1 within much less mile. It doesn't make much sense as there's still too few data for 'autonomous' car to make comparison, but based on data human driver is not so bad as some may 'feel'. And we can improve this w/out introducing many new risks, by investing in automated accident & violation prevention, aka Level 2 self-driving, and it's more efficient as the situation is relatively limited. But decision maker know only doing that doesn't make big $, and yes, that's not so dreamy. Other ppl appear to be blinded by the fact 'AI' is generally good at what it was programmed for, often better than human, which results in misbelief that they work in every situation once matured (that never happen unless they switch to completely new architecture).

 

Obviously their 1st priority was not safety. Fatal incident would be just a matter of time even if this didn't happen.

Yeah, I also felt the video strange, even doubted a bit maneuvering by police to avoid criticism to 'autonomous'car. lol

 

"Nvidia suspends self-driving tests globally

(Reuters) - Nvidia Corp said on Tuesday it will suspend self-driving tests across the globe after Uber Technologies Inc [UBER.UL] fatality last week, according to a source at the GPU conference in San Jose, California...

The chipmaker is testing self-driving technology globally including in New Jersey, Santa Clara, Japan and Germany..."

https://www.reuters.com/article/us-...f-driving-tests-globally-source-idUSKBN1H32E0

 

"Uber settles with family of woman killed by self-driving car"

https://www.theguardian.com/technol...th-family-of-woman-killed-by-self-driving-car

Well, that didn't take long -- Keep the details, cause under wraps.