well...a thesis not only has to be theoretical, it also has to be reasonably relevant to the research that your advisor is taking on. So a REALLY good first step is to talk to your advisor about his/her research interests and ask if there are any research ideas floating around their head to take on. I've seen way, way too many students try to strike out on their own without talking to their advisors first only to either come into conflict later on or choose a project that isn't acceptable.
Now, assuming your advisor is on board for animatronic heads, you might want to focus more on the artificial muscle part of the head, since it would have the most far-reaching relevance to fields outside of animatronic heads :-P I'm in grasping research at the moment, and I've been looking for a useful artificial muscle that is electrically powered (that is, not pneumatic). I'm thinking it's probably possible to construct such a muscle using a motor along with a transmission system of some kind into a compliant medium...but that's just me :-P (and yeah, I do consider it to be fun haha)
Other ideas would be a facial expression learning/mimicking robot, but you'd need the head to work with first. That might be fun, and would have a lot of consequences external to heads again -- specifically, your research would focus on learning social cues from body language, and also learning complex control of the facial muscles. (keep in mind that facial expression recognition is a pretty big field, so your advancements would necessarily be smaller)
Read the literature! You seem to be saying: "I want to work with animatronic heads". In research, what you'll want to do is start with a platform (the head) and see what you can DO with it. So you mentioned you like learning and neural nets (I have my own opinions on that :-P ) meaning you'll likely need to get the head pre-fabricated, learn to use it FAST for a masters thesis, and then set up HOW exactly you're going to use learning algorithms to do...whatever. The other option would be to build the head yourself, at which point your research would focus on making the head design better in some way, or addressing an open problem in design or control. The literature will point to where the open problems lie.
Does that help?
MIKE
ps: oh, and TALK TO YOUR ADVISOR first