What I don’t understand is that – with apologies to Clint Eastwood – there are two kinds of people: those who say self driving cars work, and those who say they only just barely work (and the technologists in the latter group say self driving cars will probably never come because the job is so complex).
I have never driven a Tesla but have been in lots of them – in some countries they are used as taxis.
Wouldn’t it make sense to take the Car-Thread out of this political Thread?
Difficult, because there is so much of this in the electric propulsion thread, and now we have self driving (working or not, etc).
there are two kinds of people: those who say self driving cars work, and those who say they only just barely work (and the technologists in the latter group say self driving cars will probably never come because the job is so complex).
China is part of the first group, with a fleet of driverless robotaxis already operational in Shenzen.
Peter wrote:
What I don’t understand is that – with apologies to Clint Eastwood – there are two kinds of people: those who say self driving cars work, and those who say they only just barely work
Some of this is due to mixing up the discussion on Tesla with a general discussion on self driving car. Although Tesla marketing and Mr. Musk often try to hint something different, Tesla is not a self driving car. They do not claim to be one (beyond some marketing speech). One of the challenges in the discussion ist that Tesla call it’s suite of driving assistants “Autopilot” that seems to be understood as more than they are – well, at least to people who are not pilots because we understand very well that an “Autopilot” is something completely different from an UAV…
Autonomous driving in general works under certain conditions – if you look at what AutoX is doing in China one can realize it takes much more sensors to make it happen than a Tesla has. To some extend (and that has been discussed widely on AI conferences) the early Elon Musk mantra “There’s nothing you can not do if you just have enough camera images” of which he refuses to admit he’s been wrong is actually one of the biggest roadblocks for Tesla to get into true autonomous driving.
Malibuflyer wrote:
That, however, might say more about company valuation methods and the state of current stock markets than about the quality of Tesla as a company or their cars.
absolutely agree with this. As with anything that is publicly listed.
What are the main challenges with self driving?
Can it even be done totally optically i.e. visible spectrum images, like a human driver uses? It ought to be possible – because humans do that. But humans have a computer behind it which is so complex it is nowhere near being understood by humans, and it has had years to learn how to interpret images.
In my business (electronics) I see a load of effort going into short range radar.
New thread created as per request
Peter wrote:
Can it even be done totally optically i.e. visible spectrum images, like a human driver uses? It ought to be possible – because humans do that.
First of all: Humans don’t do it! Humans rely on visual clues as well as on hearing and the built in “accelerometers”.
But more to the point: The problem with computer vision for autonomous driving is, that the visual environment of the traffic space is optimized for humans. And while HIs get rapidly better as interpreting images, they have still (and for the foreseeable future) a very different strength/weakness profile than humans. Humans are still much better at filtering out irrelevant information (while HIs are better of keeping track of multiple objects at the same time).
A task as simple (for a human) as identifying a red traffic light is still a huge challenge for AIs (as there are many red lights in the traffic space). At estimating distances from 3D vision humans are still much better (to be precise: Humans are much better at quickly judging if an object is within the “danger zone” or not). A task as easy (for a human) as to judge if it’s raining or not and therefore if you have to turn on the wiper seems to be almost impossible these days for AI – at least Tesla seems to have given up on it.
For all of these reasons it is state of the art for (Semi-)autonomous driving to not only rely on camera images but also have special sensors (e.g. rain) and LIDR for distance measurement. It will be interesting to see (from an AI community standpoint) if Tesla will eventually get out of the box of “camera only philosophy” they have maneuvered themselves into.
In the early 2000s I started to dislike what was happening to new motorcycles, after decades of enthusiasm since age nine when I’d bought and ridden my first for $25. Starting in the late 90s I’d tackled some of the technical problems with the new ones and made some money doing so, but decided I didn’t want to go any further with it. I’d lost my interest in new cars earlier, maybe in the early 1990s. In 2002 I decided to rekindle my long since dormant interest in planes instead, spent a fraction of that money to buy one and resumed flying – it had become apparent that instead of buying pointless, unwanted gimmickry this would be a great way to move towards the kind of sustainable engineering design I like most. I’d always planned on going back to flying some day anyway and getting away from the kind of nonsense described in this thread, self driving cars etc, was a great move for me.
To this day my newest of my 9 motorcycles was built in 2002 and I still love my plane for having owner-centered design that promotes independence, not planned depreciation and slavish dependence on infrastructure. I’ve met a lot of smart and capable people with similar interests, unlike in the motorcycle scene where buyers have increasingly become uninvolved dilettantes, paying money to be on somebody else’s depreciation treadmill. Cars have become throw away appliances for me. I buy the ones that offend me least, drive them and eventually ‘toss them in the trash’ without remorse. Actually I typically sell them to a kid for 1/10 of what I paid. In the meantime I pay as little attention to them as I can, and my main joy in my current car is that I was able to buy one with a manual transmission at a discount!
Related to self-driving cars, I wonder if they made any testing on roads covered with snow? I started thinking about that because my car has lane departure system which relies on the sensor detecting the lanes’ lines. During the last few weeks, with quite a few roads covered with snow, the system could not detect any of these lines!
I wonder how the ‘auto-pilot’ in Teslas copes with that… Humans does not have much issues with this, which is a good thing :-)
This probably deserves a thread of its own but its appears good ole Elon now controls more than 25% of all active satellites. Scary stuff… https://www.independent.co.uk/life-style/gadgets-and-tech/elon-musk-starlink-satellites-spacex-b1794888.html
Peter wrote:
there are two kinds of people: those who say self driving cars work, and those who say they only just barely work (and the technologists in the latter group say self driving cars will probably never come because the job is so complex).I have never driven a Tesla but have been in lots of them – in some countries they are used as taxis.
European Teslas are limited by UNECE regulations and can’t do much. We can watch American FSD beta videos to judge how it’s going and it’s going better with each release.
(Car is driving whenever the blue steering wheel is lit)
