I notice that r2scan could be more accurate than rscan, but I am unsure if r2scan could describe the d electron localization without Hubbard+U? Or do I still need Hubbard+U with r2scan?
If I do understand correctly your question, you ask if an exchange correlation functional (r2scan or another functionals in the future) can describe correctly strongly correlated systems.
The answer is NO as a consequence of the fact that this U is a correction of a many-body problem while DFT solves merely one body problems including corrections as nonlocal corrections. But pathological correlation effects cannot be cured that way.