Looks like some great problems (ranging from "Infinite Mario" to a helicopter simulator) to work on for this 2009 reinforcement learning competition. If you have experience or want experience programming AI, this could be a bunch of fun. I'm too busy with a mountain of other things, or else I'd probably look at that octopus problem. I just love high-dimensional action spaces.
Side note, looks like they use Java in their software, but not exclusively. I haven't dug deep enough to see everything going on.