[Rhapso]

Sensors/Awareness of environment is where the dream dies. Not just in that sensors give very limited input, but that limited input is far more than you can practically use effectively without a team of PhDs and a few years. You end up making very greedy decisions about behavior based on sensor readings, which work great in a narrow context of places and behaviors, but generalize poorly to human spaces. UAVs are a little better because their "specific context" is a bit more broad than most (in the air, minimal turbulence, and nothing to hit). If you want a similar interesting problem where you could feel efficacy look at robotic blimps or boats, you can get kits!

[Symmetry]

Robots in human spaces can be done and, more to the point, the company I'm working for is doing it[1]. And there's the Google car too. It does require spending more on sensors than a hobbyist will want to spend and in our case it requires the occasional remote intervention when, for instance, someone leaves a box in the robot's parking spot.

[1]http://www.vecna.com/on-demand-delivery

[Rhapso]

I'm definitely approaching this from a hobbyist's point of view. My experience is limited to FIRST and a thrust vectoring bi-copter that was deemed too likely to maim or kill to be used in public spaces.