Wow, I'm sincerely impressed - the Robotics Institute of Carnegie Mellon U. has jumped a pretty significant hurdle recently in "cognitive scene recognition" - they realized that by synergistically intersecting camera viewpoint, object classification, and surface geometry recognition, has finally overcome some of the most significant hurdles of robotic visual data processing.

So, basically, now visual systems don't have to worry about getting confused by "normal" things which our brains differentiate between daily (for instance, it won't confuse a window midway up a building from a car because of similar geometry, etc, because it actually takes it's position vs. the ground level into account before running it past object classification...

I'm honestly not too suprised by this news, however, and it frankly makes me scratch my head like, "How were they figuring it was going to work before they figured this out?" I'm happy, however, that these cognitive scientists have finally made the full leap into the three-dimensional universe.

But I have to give credit where credit is due; thanks to them, future robotic systems will now thankfully avoid thinking cars are smashed into buildings, and are one step closer to being able to actually smash cars into buildings because they can now discern the difference between the two sets of parameters in context with each other...

Coupled with Toyota's soon expected launch of mass-produced "personal robots", we should soon be actually seeing these little buggers trucking around on our streets (w/i the next 5 years or so)...

Now the question is; when you get one, will you put a lowjack system in it to keep "tech gangs" from ganking your hardware or just keep them around the house to do your laundry? ;)