SoylentNews is people
SoylentNews
from the but-officer-my-car-meets-government-standards dept.
According to this news item, one of the holdups to the wide usage of autonomous cars is a lack of regulations that can be used to certify the control systems: http://www.automotivetestingtechnologyinternational.com/industry-blogs.php?BlogID=1973
Regulatory challenges
The influence of legislation around the world on the way tests will be performed in future is relatively small. For example, there are as yet no binding standards for driverless cars. This makes it a very complex task to make cars reliably safe for the global market. However, there are of course calls for safety levels, backed up by defined safety standards such as ASIL. They are a pre-requisite for planning reliability for investments in necessary new testing equipment. Crucial for the breakthrough of autonomous driving will be the speed at which global legislation can introduce the appropriate regulations. The sooner this happens, the faster the requirements for validating a completely driverless car can be implemented.
(bold added by submitter)
The same author suggests that the well established V-model for system development, validation and verification might be short-cut in some way to meet aggressive timing requirements -- which sounds like a great recipe for disaster to this AC. Have any SN readers had any involvement in this area?
A general reference on V-models is an interesting read. According to the article, it started at Hughes Aircraft in the 1960s (Los Angeles).
(Score: 2) by archfeld on Saturday May 20 2017, @02:41AM (3 children)
You say that like there aren't grizzly auto accidents that occur everyday with human drivers behind the wheel. Honestly other than in a few odd conditions I think that the severely decreased reaction time of autonomous vehicles will cause a noticeable decrease in collisions, especially in single vehicle situations.
https://www.google.com/#q=percent+of+car+accidents+caused+by+human+error&spf=1495247801066 [google.com]
For the NSA : Explosives, guns, assassination, conspiracy, primers, detonators, initiators, main charge, nuclear charge
(Score: 3, Informative) by kaszz on Saturday May 20 2017, @02:03PM
There are, but autonomous car will be unpredictable. They can go mad at the same time in droves. They can be used for remote attacks. And so on.
(Score: 2) by AthanasiusKircher on Saturday May 20 2017, @04:35PM (1 child)
I doubt that fatal accidents from AI cars will be treated by the media, regulators, Congress, etc. the same way we treat human drivers. I have no doubt that what you say is correct -- and even with the current state of AI tech, we'd probably save lives compared to the average human driver.
The questions are going to emerge when some horrible accident occurs with AI because -- especially if it's "grizzly enough" or unusual enough -- there will be questions about whether it could have been preventable. Some of that may just be humans imagining that they could do better when they couldn't. (The lack of control does odd things to risk perception; hence why many people fear flying, but don't fear driving, though the latter is far more dangerous -- in driving, though, they are "in control.")
But some of it may end up questionable if an accident is due to a sufficiently weird "edge case" that AI likely couldn't handle. At that point, it won't matter if the AI car overall is more safe statistically: media perception will likely be that the car "caused the death" of passengers and/or others in a case where a human driver could have performed better. it could create a consumer or regulatory panic that leads to avoidance of AI cars until they are proved to be more "secure."
(Score: 2) by carguy on Saturday May 20 2017, @05:13PM
While I agree with most of your post, this bit:
> ... we'd probably save lives compared to the average human driver.
seems to keep coming around.
I'm not an average driver, and I'll guess that you are not either. I'm middle aged (no teen hormones), I don't drive drunk or otherwise impaired, don't work two or three jobs (implying driving when extremely tired) and I'm usually not in a hurry (plan my day to avoid the need to rush). I'm going to wait until the self-driving cars have stats that look as good as people like me.
Separate point -- here's an interesting take on the relative safety of flying vs driving -- based on number of trips, not on passenger-miles,
http://www.science20.com/gerhard_adam/flying_or_driving_which_safer [science20.com] Here's a cut from the end of this approach:
If we take some arbitrary numbers, we can begin to approximate how these numbers relate. With 10,090,274 domestic flights in the U.S., that becomes the number of possible events that could've resulted in accidents. With cars, I'm going to make the following assumption that taking all the licensed drivers (202.8 million) and assuming that they only drive to and from work. This would result in two trips per day for about 261 days or 522 trips per driver. As a result, the total number of driving events would be a staggering 105,861,600,000 events. On this basis, we can see that an automobile driver is 10,000 times more likely to be exposed to the circumstances leading to an accident.
Driving looks pretty safe when seen this way.