Except it doesn't, because there's a massive amount of data radar, especially automotive grade radar, can not supply.
Redundancy means it can stand in for another component. Radar can not stand in for vision. They do different things in fundamentally different ways.
Radar was making the system
worse
Again all this is explained, with real world data showing why radar was removed, in the AI video you keep refusing to watch.
Here's Elon mentioning it again earlier today-
You're re-posting assertions and claims by Musk that are unsubstantiated by data. Every other manufacturer (except maybe Hyundai or somebody) has done years of testing and determined that multi-sensor fusion is needed and that cameras-only cannot provide the performance and safety under all conditions that they seek.
You keep telling people to watch the Tesla propaganda "AI day" video. Watching Tesla showcase their internal research efforts and make claims without any data or independent review is not an objective way to answer these questions. The fact that you don't seem to understand that "AI Day" was a *promotional* exercise and not an independent assessment of their technology is really disturbing, honestly.
The claim that radar was "making the system worse" is unsubstantiated. It may be that Tesla was doing multi-sensor fusion poorly compared to other manufacturers, but that's on them, not on radar.
Moreover, you're refuting your own claims when you say the sensors "...
do things in fundamentally different ways...". Yes, that's the point. You
need different types of sensors and inputs that work in different ways in order for the system as a whole to work well under different real-world circumstances.
Radar directly measures range (distance). Camera-based distance measurement attempts to *estimate* distances from features and across frames in cameras, and is prone to all kinds of error sources that radar distance measurements don't suffer from. And before you try to explain to me how these sensors work, understand that I'm a physicist who works for a major aerospace company that literally does multi-sensor fusion as our business. I know the differences among radar, LIDAR/LADAR, and cameras and why one uses different combinations of these under different circumstances.
In cars, specifically, it's been shown that vision-only systems perform much more poorly at night in the presence of pedestrians and bicycles, as just one situationally-dependent example. There are others.
Here's one of many studies you can go read, I won't go do an entire literature search for you, but this is a good start:
https://www.sciencedirect.com/science/article/pii/S0749379718320932
Moreover, radar HAS NOT BEEN DEACTIVATED in Teslas that have them, even with "Tesla Vision"...it's still used by Tesla's software loops for automatic emergency braking (AEB) and forward collision warning (FCW)...why do you think that is, if radar doesn't add value? Weird, huh?
In fact, there's some evidence this is true even in so-called "FSD" vehicles *when* FSD is activated. They're still checking radar when it's available. LOL.
In addition, with the change to "Tesla Vision" in some Model S vehicles running 2022.20.9, the "1" setting in the cruise control distance keeping has been removed. If "vision only" is better, why did they eliminate that setting? The answer is because they no longer have direct radar range measurements and don't want to let people follow that closely anymore because "cameras only" has some limitations. It's a tacit acknowledgment on Tesla's part that "vision only" doesn't work as well for that application without direct radar range measurements.
In addition, Tesla just registered a new, high-resolution radar in an operational physical package with the FCC. Why do you think that is? Below are a few source links, which includes the following Elon Musk quote:
"A very high resolution radar would be better than pure vision, but such a radar does not exist. I mean vision with high res radar would be better than pure vision." - Elon Musk
Except, Elon, very high res radars *do* exist, and that's the one Tesla just registered with the FCC a few months ago with an automative-grade, vibration resistant case that appears ready for production use...LOL.
The fact is that Elon had Tesla remove radars -- and forego LIDAR -- to keep costs down and, later, deal with pandemic supply chain issues.
He's also trying to get around having to make his team improve their multi-sensor fusion knowledge. It's unfortunate, because they were sort of doing just fine with radar/cameras early on, and were slowly eliminating things like phantom braking, and other issues with how they did multi-sensor fusion.
The whole Tesla game of using "neural net" training and camera-only automation is an attempt to force the inferior cameras-only method of automation to work. It will not work as well as multi-sensor systems, and Tesla will eventually put the new radar they registered in their vehicles...possibly sooner than you think.
Tesla has filed to use a new radar in its electric vehicles, and everyone is confused since Elon Musk has...
electrek.co
Tesla registered a new high-resolution radar unit with the U.S. Federal Communications Commission (FCC). Is radar coming back?
cleantechnica.com