The military's DARPA (Defense Advanced Research Projects Agency) has always said their goal is DARPA hard projects. That is, new technologies or systems with chances of failure (or irrelevance) that are beyond the thresholds acceptable to one of the services acting alone, or an unsubsidized contactor, or a VC for that matter.
In the last few years, DARPA scored a stunning success with its Grand Challenge series of autonomous land vehicle trials. 'Series' by courtesy, as the desert navigation task was surmounted on its second running, in what must have been one of the highest ROIs achieved in a DARPA project, even given the organizational overhead in running the $2m prize event. At the time I commented (on Winds of Change IIRC) that the real challenge would come with urban and highway driving. Well....
They're back! (PDF) Grand Challenge 3 will be urban navigation, in an environment shared with other vehicles, and with the 'bots required to obey traffic laws and conventions. (Thereby achieving a level of self-control beyond most human drivers.) There's a fun and sometimes informative slashdot thread on the Urban Challenge.
This is putting the number of variables to be managed way up, and therefore moving up the threshold for credible participation. DARPA seems to realize this. Unlike the desert challenge, this version has two tracks: A team can apply for a modicum of government funding, or raise its own and compete in trial run-offs. But all survivors compete in the same final run in November, 2007. One surmises the DARPA funded competitors will have greater obligations to grant IP to the government, but let's face it: A megabuck isn't going to go far in this project; it's an incentive to participate, but far from full funding for a fresh start.
Like the previous Challenge, this one isn't aimed (directly) at building fighting 'bots. This portion of DARPA's autonomous control work has been explicitly targeted at logistics functions: reduce the number of humans exposed to ambush and IEDs while moving supplies, and increase the 'tooth to tail ratio' of the total force. However, it doesn't take too much thought to come up with systems that could be dropped onto an autonomous platform (long range recon bot?) or novel operational ideas (logistics elements that launch across the desert before the human strike force?). Even if the systems aren't sufficiently robust for full autonomy, a strong self-operation platform would allow human supervisors to run large spans of cooperating vehicles and weapons systems. Robotics is one of the spaces where war-driven innovation is accelerating. Keep watching this spot.