I find this page of NewScientist.com very interesting. You can find videos of several fascinating robots in the page. Among those that I have seen (not all of them), I found 1) the self-replicating robots of the Cornell university (see Hod Lipson’s page), 2) dancing Qrio robots of Sony, and 3) Asimo of Sony more interesting.
This post is a great place to ask this question: What are remaining open problems in robotics?
I know there are many difficult problems in the field. However, is there any problem that we are far behind solving it? One answer is, of course, the problem of having intelligent machines. Putting it aside, is there any mechanical/control-oriented difficult challenge?