I don't know much about the inner workings of Asimo, but the other robots are good examples of the typical extremely shallow approach to problems involving perception and understanding. Superficially the videos of LUDWIG look impressive, but this just involves the robot tracking some brightly coloured blobs against a completely unchallenging monotone background. This kind of coloured blob tracking has been around for well over a decade, and its simply not scalable to real environments or real problems (you couldn't take the robot out into the street and have it behave with a similar level of competence with real objects). These types of perception system carry a heavy legacy from the world of industrial vision systems, where lighting is carefully controlled and objects presented in a very predictable manner.

I'd like to see robots tackling the more difficult perception problems, so that they may result in truly practical applications in everyday situations.

I think all of us naturally gravitate to the robotic projects that are trying to emulate human perception; such as object tracking through machine vision.

But I wonder if the first impressive real world navigation and object identification systems for robots will come instead from the upcoming implementation of sensor nets.

The combination of GPS, enhanced data from sensor nets, and simple radar or sonar capability could give a robot a lot of information to be used in navigating a real world situation.