Tesla and Uber Unlikely to Survive (Vol. 2)
Discussion
Tuna said:
I've highlighted the big issue here. Even with LIDAR, cars are using 'offline' maps of areas, which means they're basing decisions on something that may be completely out of date, or just wrong today (a sudden contraflow system, or a burst water main, or even something as simple as someone crossing unexpectedly from the other side of the road). This is the point where the human ability to interpret an unexpected situation exceeds a computer model every single time.
You're absolutely right that with 100% accurate information, cars can drive themselves more accurately and more safely than humans. The problem comes when that information is not 100% accurate, and when the humans around you are doing, well, human things. Your machine driven car brakes to a perfect halt when it sees an obstruction, and the human driver in the car behind it drives straight into the rear end. And the scale of the problem means 10000s of miles without disengagement are not enough, when the cumulative car travel each day is measured in millions of miles just for the UK.
I'm happy to believe the problem will eventually be cracked, but I don't think it's nearly as soon as people would like to think.
So you agree tesla's vision first approach is the better long term strategy over mapping? Though tesla use maps for speed limits, stop signs (!) and bridges, so I'm not saying they are close to that yet. You're absolutely right that with 100% accurate information, cars can drive themselves more accurately and more safely than humans. The problem comes when that information is not 100% accurate, and when the humans around you are doing, well, human things. Your machine driven car brakes to a perfect halt when it sees an obstruction, and the human driver in the car behind it drives straight into the rear end. And the scale of the problem means 10000s of miles without disengagement are not enough, when the cumulative car travel each day is measured in millions of miles just for the UK.
I'm happy to believe the problem will eventually be cracked, but I don't think it's nearly as soon as people would like to think.
Re rear ending. Sounds to me like you are arguing for more automation, not less? Shouldn't the human and robot alike, both stop for the obstruction? The inability of the driver behind to stop in time, seems like a current problem that needs to be fixed. How many humans rear end other human cars every single day?
Tuna said:
sambucket said:
I think people overestimate how hard the cognitive aspect of FSD is, and underestimate how difficult the 3D visualisation is. Once you have accurate 3d real time mapping, it's not that hard. As fully kitted out lidar test cars with HD maps have shown, you can travel 10000s of miles without a disengagement if you have every inch of route mapped out in 3D.
Also, the human aspect is perhaps overplayed. If a car can react and adjust in an instant, you don't need to know the emotional reasons for another driver making erratic decisions. You just need to be able to react in time.
I've highlighted the big issue here. Even with LIDAR, cars are using 'offline' maps of areas, which means they're basing decisions on something that may be completely out of date, or just wrong today (a sudden contraflow system, or a burst water main, or even something as simple as someone crossing unexpectedly from the other side of the road). This is the point where the human ability to interpret an unexpected situation exceeds a computer model every single time.Also, the human aspect is perhaps overplayed. If a car can react and adjust in an instant, you don't need to know the emotional reasons for another driver making erratic decisions. You just need to be able to react in time.
Edited by sambucket on Thursday 9th July 23:21
You're absolutely right that with 100% accurate information, cars can drive themselves more accurately and more safely than humans. The problem comes when that information is not 100% accurate, and when the humans around you are doing, well, human things. Your machine driven car brakes to a perfect halt when it sees an obstruction, and the human driver in the car behind it drives straight into the rear end. And the scale of the problem means 10000s of miles without disengagement are not enough, when the cumulative car travel each day is measured in millions of miles just for the UK.
I'm happy to believe the problem will eventually be cracked, but I don't think it's nearly as soon as people would like to
sambucket said:
So you agree tesla's vision first approach is the better long term strategy over mapping? Though tesla use maps for speed limits, stop signs (!) and bridges, so I'm not saying they are close to that yet.
Not really - vision brings its own set of problems, and we've learned over the last couple of years that the machine learning equivalent of 'optical illusions' are very subtle and unpredictable. The engineer in me feels instinctively that Lidar at least gives you a robust measurement of 'hard surfaces' around you, but really in both cases it comes down to how reliably we can infer information from an incomplete data source.sambucket said:
Re rear ending. Sounds to me like you are arguing for more automation, not less? Shouldn't the human and robot alike, both stop for the obstruction? The inability of the driver behind to stop in time, seems like a current problem that needs to be fixed. How many humans rear end other human cars every single day?
Yes, in many ways the most difficult part of the transition may well be the point where we have an odd mix of autonomous and human driven vehicles.The rear ender issue swiftly moves into the trolley car problem. If there is a sudden unexpected obstruction in front of us, what do we do? Take to the opposite side of the road, drive on the pavement, brake as hard as we can, or brake as slowly as possible to reduce the chances of a rear-end shunt? I had a very lucky escape many years ago when an over enthusiastic overtaking maneouver met some unexpected oncoming traffic (incomplete information!). I ended up in a ditch rather than trying to force myself back into the lane - and stopped miraculously parked between two very solid trees.
The thing is, there will be some classes of accident that autonomous cars are infinitely better at avoiding than mere humans. But equally there will be others that they are worse at, or handle in a way that causes more harm. We've not really got to the point where we can start thinking about what that means for us, and for regulations.
hyphen said:
Good timing, Elon gave a speech today:
https://www.dailymail.co.uk/news/article-8506589/E...
People didn't believe him when he said he would land rockets. When he did it they doubted they would be reused. Now seeing two rockets carrying out a synchronised landing is a matter of routine.Elon said:
'I remain confident that we will have the basic functionality for level 5 autonomy complete this year.'
So what exactly does 'basic functionality' mean? Politicians answer!https://www.dailymail.co.uk/news/article-8506589/E...
Unlike politicians he does deliver on what he says, it's just the time lines he gets wrong.
98elise said:
People didn't believe him when he said he would land rockets. When he did it they doubted they would be reused. Now seeing two rockets carrying out a synchronised landing is a matter of routine.
Unlike politicians he does deliver on what he says, it's just the time lines he gets wrong.
All this crazy long thread has ever been is 'Tesla/Musk won't do that'. Tesla/Musk do it. 'Tesla/Musk might have done the last thing, but they will never do the next thing'.Unlike politicians he does deliver on what he says, it's just the time lines he gets wrong.
Currently things they won't do: CyberTruck, Full self drive, the Semi (going a while back people were saying it was impossible).
We will probably see the CyberTruck and the Semi within the next 12 months.
I come here to see the tears and moving of the goal posts because it's comedy gold.
Really all the doubters have left is FSD now. Can't wait.
RobDickinson said:
Cheeses of Nazareth said:
Sadly, the world is not testable and predictable.
I will take my chances with someone who has a dog in the fight
or Karen who can barely tell which pedal makes the car go forward?I will take my chances with someone who has a dog in the fight
What has that got to do with anything. Is Karen, who is obviously too thick to perform basic tasks , able to earn enough for a self driving car .
jjwilde said:
All this crazy long thread has ever been is 'Tesla/Musk won't do that'. Tesla/Musk do it. 'Tesla/Musk might have done the last thing, but they will never do the next thing'.
The only technical thing I've said he will have problems delivering has been FSD. I think a full, ground up re-write of the software, and a complete hardware revision has shown that, yes, it is difficult to deliver FSD and that his early claims were in fact false.Claiming it's "just a delay" ignores the fact that his original statement was that he would deliver FSD on a system that has since been completely replaced (new hardware and new software).
Jam tomorrow.
Lighten up Rob!
ked off to warmer climates. 
Gassing Station | EV and Alternative Fuels | Top of Page | What's New | My Stuff