For the moment, DARPA - that Shiva-like organization that has provided such wonders as the internet and such horrors as killer drones - insists that its humanoid robots are designed solely for support roles and for testing equipment designed for use by human troops. And as with ARPAnet, these robots may some day be more prevalent in benign, Asimovian roles than as bringers of death and horror, but for the moment we can put off both the nightmares and the fantasies and simply marvel at what they're doing. Videos below the fleur-de-Kos.
First, here is a video showing the evolution of two humanoid robots, Petman (Protection Ensemble Test Mannequin) - a humanoid robot ultimately designed for testing troop equipment - and Atlas, from relatively simple bipedal walking machines in 2009 into something much more science fiction-y today, and also briefly shows some non-humanoid robot projects:
The humanoid robots are clearly entering the Uncanny Valley where a "cute" resemblance starts to become disturbing. Atlas in particular has built on the lessons of Petman and evolved into something that is actually more frightening than uncanny, as you can see in this video of its evolution (you can skip the animation in the first part unless you're curious about the technicals):
There will never be a future exactly like either Asimov's novels or the Terminator movies. First of all, most things that a machine can do better than humans are better handled by machines specifically designed to do that one thing rather than a general-purpose robot. Also, even if they're used as Skynet-style killers in war or even as enforcers domestically, that doesn't preclude benign applications, nor vice-versa. There may come a day where major conflicts are generally automated and human involvement is mainly to die as helpless civilian casualties while the robots duke it out Transformers-style. I don't know whether that's worse or better than Man's inhumanity to Man - maybe it's both.
Of course there's always the Frank Herbert side of things vs. the Asimovian: That robots would simply empower their human owners to enslave those who don't have any robots. Herbert always was keener on political reality rather than the Gibbon-esque historiography of Asimov. They could both be right: A society like Asimov's Spacer culture - rich, complacent, totally dependent on robots - could evolve, but he left out something that Herbert did not in his fictional history: That the owners of robots would first have to exterminate the surplus humanity they no longer needed to serve them, which would be growing increasingly restive as their economic prospects dimmed.
None of this is inevitable though. Human beings are not passive tools of economics: We make conscious choices when we first make the "choice to choose," so we can choose a future where the growing productivity and comfort of an increasingly automated society is widely distributed and creates a post-scarcity world. That will have its own dangers and limitations, its own disasters and horrors. So when I see these DARPA videos, they are frightening, but within the fear they engender is the excitement of benign opportunities and possibilities.
DARPA is one of those awesome things that represents a cynical excuse for a humane purpose, parasiting off of an institution that is itself parasitic (DoD) in order to return some value back to humanity. I find that philosophically amazing and inspiring.