Wol747
Active Member
Very inconsistent. Today I was in an 80 Km/Hr area and came against a road work 40Km/Hr sign: the car recognised that but just carried on with it after the zone until it saw another 80 sign miles down the road.
You can install our site as a web app on your iOS device by utilizing the Add to Home Screen feature in Safari. Please see this thread for more details on this.
Note: This feature may not be available in some browsers.
When it comes to reading signs, AP2 has been way behind AP1 for years. Elon keeps thinking "who knew speed signs were that complicated!?! Oh yea, those MobileEye guys knew"AP1 uses an entirely different software stack.
AP3 certainly reads speed limit signs, partially with 2020.32 and now fully with 40, when it comes to static speed limit signs in USA, at least the ones I've experienced.
Tesla's sign reading problems is a failure of deep learning.
So is the issue of cars missing from the display when stopped at a light. Getting most of the desired functionality fairly quickly is a characteristic of deep learning. So is hitting a brick wall regardless of the quantity of additional training. Tesla has apparently not achieved the advances in deep learning that they needed to achieve FSD.
Yeah...there’s a VERY long tail on this thing that the machine learning will not get. I have close to zero confidence that full self driving without requiring a drivers hands on the wheel is coming anytime soon.Tesla's sign reading problems is a failure of deep learning.
So is the issue of cars missing from the display when stopped at a light. Getting most of the desired functionality fairly quickly is a characteristic of deep learning. So is hitting a brick wall regardless of the quantity of additional training. Tesla has apparently not achieved the advances in deep learning that they needed to achieve FSD.
The car has to handle 100% of situations right 100% of the time. Too many crazy scenarios and variables in the real world for the computer to understand them all.
Yeah...there’s a VERY long tail on this thing that the machine learning will not get. I have close to zero confidence that full self driving without requiring a drivers hands on the wheel is coming anytime soon.
The car has to handle 100% of situations right 100% of the time. Too many crazy scenarios and variables in the real world for the computer to understand them all.
The difference there is the humans take the liability and are held responsible. Is going to accept liability and responsibility for every mistake their cars make? Are regulators going to ALLOW Tesla to take that liability if they’re willing? You can understand the difference here, right?Humans don't. Autonomous vehicles have to be measurably better than humans almost all of the time and then try to handle the other situations as safely as possible.
Yep. I work for a large publicly traded software company. We employ a number of PhD data scientists that specialize in machine learning, image recognition, spacial analysis, etc...When Tesla started their FSD other sophisticated companies were very optimistic about the reach of machine learning. The voice assistants are the best example. Even Apple was very optimistic about Siri in 2016. All these companies hit the wall where more data doesn't improve results.
Tesla experienced the unfounded optimism in the now famous "3 months maybe 6".
So what Musk is today claiming is that Tesla have broken through this machine learning barrier. Yet he can't know, because they have not proven a large improvement in perception.
Today it is hard to be nearly 100% certain to simply determine if a human is present in a video image. Security camera systems struggle with this seemingly simple functionality. If Tesla has made a breakthrough a great many products can be improved with better machine learning.
Here in my rural area, there is a sign that says "50 km/hr Ends". I know that that means I can now go 80 km/hr, but my AP1 Model S does not, and remains at 50.