Depends on the structure of your Msc, the funding, the compute available, etc, you need to give those details if you want a detailed answer. If you choose something at the limits of the compute available to you, then you make things 10x as difficult.

The easy answer that applies everywhere is, choose a topic that has a good volume of phds, postdocs, or professors at the college you're joining. The deep knowledge that arises from this no matter which subfield in RL it is in will later help you transition to your desired subfield much better than a surface-level effort directly in your desired subfield.

The most _likely_ answer is sim2real. It has just the right mix of generative AI (funding), robotics usecases (there is an easy "end" for your thesis to hook onto as the "impact"), and lots of unexplored paths (you won't be chasing a common frontier competing with hundred others). But like I said this depends on the previous two points.

Of course there's what you're passionate about and so on, but at the end of the day objectively deep knowledge matters more than passion, unless you're so deep into something, but then your choice is made and you wouldn't have asked this question, so prioritise deep knowledge. Try to actively reduce breadth, IMO the purpose of a research-oriented program is to learn to study depth-wise to the end.

As for RL in BCI, as far as I know, we still use established "RL 101" algorithms directly applied to the brain signals. Any research deep enough (but short-term enough for a MSc) here will be less RL theory/algorithms and more about the details of this specific application. If that's what you want, go for it.