Posted by mhb 4 days ago
Note: I have not had the pleasure of riding in one yet, but from what my friend in SJ says, it’s very convenient and confidence-inspiring.
The drive was delightful and felt really safe. It handled the SF terrain, traffic and mixed traffic like trams very well.
I wouldnt trust a self driving tesla ( or any camera only systems) though!
Drivers can and do misuse adaptive cruise control systems, sometimes with fatal consequences. Memes aside, there is no strong evidence that fatal misuse occurs more frequently by owners of Tesla cars than with comparable systems from other brands.
This perception reflects the Baader–Meinhof phenomenon, more commonly known as the frequency illusion. Nobody is collecting statistics for other brands, so it’s assumed the phenomenon doesn’t occur.
A similar pattern occurred with media coverage of EV fires. Except in this case, good statistics exist which prove the opposite: ICE vehicles catch fire more often than EVs.
I own a Tesla and paid about $10K for the full self driving capability a few years ago. Yeah, I would not trust a Tesla to drive me from airport to my house. There is a reason Tesla is still stuck at level 2 autonomy certification and not 3, 4 or 5.
There aren't a million Teslas with FSD active in the US. According to Tesla in their latest earnings report there are 1.1 million people worldwide with FSD.
There's no way a sensor can tell if a signal was from its origin?
Guessing any signal should be treated as untrusted until verified but I suspect coders won't be doing that unless it's easy
I mean it doesn't. If you actually look at it comma.ai proves that level two doesn't require lidar. Thats not the same as full speed safe autonomy.
whilst it is possible to drive vision only (assuming the right array of cameras (ie not the way tesla have done it) lidar gives you a low latency source of depth that can correct vision mistakes. Its also much less energy intensive to work out if an object is dangerous, and on a collision course.
To do that in vision, you need to work out what the object is (ie is it a shadow) then you have to triangulate it. That requires continuous camera calibration, and is all that easy. If you have a depth "prior" ie, yes its real, yes its large and yes its going to collide, its much much more simple to use vision to work out what to do.
As far as distinguishing shadows on the road, that's what radar is for. Shadows on the road as seen by the vision system don't show up on radar as something the vehicle will run into.
The SAE autonomy scale is about dividing responsibility between the driver and the assistance system. The lowest revel represents full responsibility on the driver and the highest level represents full responsibility on the system.
If there is a geofenced transportation system like the Vegas loop and the cars can drive without a human driver, then that is a level 5 system. By the way, geofencing is not an "SAE level 5" requirement. Geofencing is a tool to make it easier to reach requirements by reducing the scope of what full autonomy represents.