A discussion about the benefits of radar should account for how modern AI based on neural networks makes decisions. Understanding this clarifies Elon's comment about not needing radar and thinking probabilistically.
The NN AIs in Teslas are basically probability calculators. The decision making in NNs is probability based, not binary or rule based logic (e.g. not a rule like check under the car in front and see if another car is slowing down). NNs have weights (coefficients) applied to each input (pixel) of each sensor (camera, radar, etc) which generate outputs that are fed into more layers that have weights that effectively use input to predict what the proper objects are or behavior should be.
Determining the coefficients and NN models is hard, but it's easy to see that Tesla can assess the contribution of radar to the system. Try one pass without the radar, try it again with the radar, then compare the accuracy of predictions.
Arguing that radar handles certain cases is binary thinking because it assumes that a particular "case" is recognizable (aka TRUE/FALSE) and would handle it a certain way. However, NNs assign probabilities to predictions (e.g. 82.35% chance of being this situation), so it's not black or white.
There's also the matter of ROI when adding sensors. Using completely made up numbers, let say radar ($100) improves NN predictions by 5%, but adding two more cameras ($20) improves NN predictions by 10%. Clearly, Tesla is better off installing two more cameras instead of one expensive radar.
If two cameras improve NN predictions 10%, is that worth $20? After all, maybe accidents would drop from 10,000 to 9,000. That sounds like a lot, but autonomy will require 5 or more orders of magnitude improvement. In other words, reducing errors from 10,000 -> 1,000, then 1,000 -> 100, etc until it's .1 error. With such large reductions, 10% would not make a difference, because the 100,000% improvement needed from better AI models and training would dwarf a 10% improvement through additional sensors.
My numbers are made up, but Tesla knows what the real numbers are. If Tesla drops radar, it's because they have done the math that shows it does not add a material improvement, relative to improvements in FSD software.