If you need a workaround, your stuff doesn’t work.
7 Likes
The law of diminishing returns pretty much owns us here.
Can we make autonomous cars that hit no more bicycles than human drivers do (for a comparable number of miles driven)? Yeah, I imagine that’s doable.
Can we make autonomous cars that hit quite a bit fewer bicyclists per mile driven? Yeah, that can probably be worked out, at a cost (in dollars, design lead time or in slower drive time with more stops that turn out to be false positives).
Can we make autonomous cars that never, ever hit a bicyclist (thinking here primarily like a risk averse corporate lawyer)? In a word, no.
4 Likes
I don’t see why. People have no problem at all recognising other people. If computers can’t do that, they’re clearly not smart enough yet. They need to be trained better.
Extra sensors surely help, but people are mostly doing this on sight too. The pattern recognition systems need to be smart and broad enough to be able to recognise any obstacle, and not just the few the developers thought of while developing the system.
If I see a totally unrecognisable alien spacecraft land in front of me, I’m still going to stop, even if I have no idea what it is. If self-driving cars are unable to deal with cyclists, pedestrians, pets, children dashing onto the street, horses with or without cart, jet fighters landing on the highway, toppling trucks, or any other unexpected obstacle I can’t currently think of, then they are simply not suitable for real life road use yet.
3 Likes
Google Photos can’t tell the difference between gorillas and black people well enough to trust it with personal photos. And that’s with a high powered back-end. I think you’re underestimating the difficulties of image and pattern recognition.
3 Likes
although hopefully it would stop the car for either…
1 Like
It depends. Is it a police car?
I’m not. I’m just saying that if this is true, then it’s not good enough yet.
3 Likes
Sif the rabid pack of plutocrats who run shit won’t just ram this through as soon as they can get away with it, in a rush towards banning human drivers, before ultimately having us carry RFID ID…
Bam! Lockdown.
I welcome self-driving cars in my neighborhood. It would act as a speed break for the nimrods that currently use it to get around a chokepoint on the local throughway.
1 Like
But it’s not comparable, as humans are far better at visual processing than computers are going to be, certainly for the foreseeable future and possibly ever, and they have context and understanding of the environment. That allows them to differentiate an empty plastic bag blowing in the wind to, say, a dog, and not just identify but intuitively understand how different entities - a car, bike, pedestrian, skateboarder, etc. - are likely to move. They can put together contextual clues - e.g. a ball bouncing into a street may indicate a child will soon chase after it. They can notice a driver or pedestrian is distracted or impaired and give them a wide berth. There are a whole lot of things that AI will never do that informs driving for humans. Humans are still crappy drivers, on the whole. Machine drivers need all the help they can get. Vision alone won’t do it for them, but extra sensors add data that can give them a more complete picture of what’s going on - including information that humans can’t get through vision alone either.
5 Likes
And yet, I’m not sure if self-driving cars that lack that contextual information are really a good idea. Don’t get me wrong, I’d love not having to drive anymore. I’d love to read a book or take a nap or something. But self-driving cars need to be able to handle themselves responsibly in all circumstances they can find themselves. If they can only deal with optimal circumstances, they can only be used in optimal circumstances, and should hand control back to the driver in situations where something unexpected might happen. Which, frankly, is almost always.
1 Like
I don’t think they need that contextual information - plenty of drivers are oblivious - they can simply respond immediately to current situations. But I think getting data from more than just cameras is crucial for them to be able to react, safely and rapidly, without it.
On the rare occasions that I get out my old bike, I don’t even wear a helmet, much less a transponder… instead I work on the principle that at least some of the cars are actively trying to hit me at all times so I should be exceedingly alert.
Maybe I took too many defensive driving courses. But that car insurance reduction was so seductive!
1 Like
you need to have one on any object that could potentially end up on the road. It is also important to remember that there is a human tendency to put complete trust in such systems.
The biggest hurdle, as it seems to me, is getting through the phase where the AI is in control, but the driver needs to be able to take over in a split second. It is very hard to stay alert in such situations. So the phone comes out, or a book, and soon the driver is napping in the back seat. In those cases, there will not likely be time for the driver to take over, orient themselves to the situation, and take appropriate action to avoid collision.
This is something that people have been researching for years, in other types of transport, where the operators are presumed to be experienced, trained professionals, who are aware of the consequences of loss of vigilance.
We also have to figure mischief into the equation. Cheap, readily available objects that cars have to avoid are going to end up getting tossed onto roads everywhere. If I was a grumpy person who lived on a once quiet street, which is now buzzing with autonomous traffic, I would conceal the devices on the roadway.
2 Likes
“Today, US Airforce escorted a cyclist who had his radio beacon turned off with two F22 back to international water. A spokesperson said that due to diligence and vigilance, at no time, any danger was imminent for the national security of the US.”
1 Like
“Your honor, I have no idea how those nail-filled potatoes ended up on the road.”
RoboCar: “It’s a … ummm.”
4 Likes
I can’t imagine any AI vehicle being licensed for general use as long as this condition is in effect.
As you note, we have not solved this problem for trains and for aircraft, and there are the crashes to prove it.
This topic was automatically closed after 5 days. New replies are no longer allowed.
