Elon has mentioned that the reason they are getting rid of Radar is the conflict caused when Radar visualizes one set of data, and Vision reports a different set of data. Lots of time and processing is necessary to sort out which data set is going to be used to direct FSD commands.
This lag consumes a tremendous amount of data for logic to determine which data stream will be best used to direct your car.
The decision has been made that Vision is the most advanced and has the highest potential to effect FSD. Logic would lead them to develop a visual system most like what humans are using (two eyes) to provide the data to the CPU (brain) to direct the best path for the car.
That is why the radar is being eliminated.
Doing so will reduse cost, free up space for other electronics, reduce power consumption, eliminate conflicting data points, save weight, make frontal crash repairs less expensive, and focus all engineering talent and asset on optimizing Vision based navigation.
Yes, this is the reason, but Elon lives in an echo chamber and has misled at minimum, perhaps unwittingly because he hired many yes-men during critical stages of what he believed was vertical integration of FSD, every step of the way after AP1.
Much of this could've been avoided if Tesla paid Mobileye licensing fees for a few years, which he deemed egregious at the time, and aqui-hired its engineers when stock starting skyrocketing. Instead, he started from scratch and Intel took Mobileye for what is a trivial price now. It took years for HW2/HW2.5 to come remotely close to Mobileye's AP1 because of that hubris, the same hubris in over-engineering the MX he now wishes could only be offered in 6-seats (he has 5 kids). People are experiencing the same phantom-braking I experienced on radar in 2018 while on vision in 2022.
He promised there was enough compute-power to use vision implicitly with AP2; then, he said AP2.5 wasn't an update and only for redundancy, while privately and simultaneously hiring a legendary chip architect from AMD to make HW3 to build out sub-processing via neural nets, which has gotten things back on track for the more realistic 10-year roadmap to some kind of FSD, but look at the supposed neural nets for windshield wipers. There will be HW4 and HW5. I understand the rationale, but there's more to the story than conflicting radar/vision logic. There isn't enough compute-power. The years with Scott were terrible, and not all his fault, before Karpathy came into the picture.
The only feature from EAP I use in the Bay Area is auto-lane change. That is why it was folded into FSD, and why they are now prioritizing those orders to solve for previous mistakes: massive R&D into something not ready for normal families. FSD has made great progress recently, but I'll continue to pass.
NHTSA may put the kibosh on his plan, as phantom-braking complaints increase. It's cool watching solo dudes navigate city streets, but there is no way I'm bringing my young kids along for that experience. NOA is still less safe than my driving, and I've never had a ticket driving a car.
I realize FSD is a controversial topic on this forum, particularly because uptake is higher here, but you need to be careful to take the man at his word at every step of the way. He is an unfocused CEO w/o a COO. He needs his Tim Cook.