It's pretty cool, but even cooler if they used also with their computer vision techniques the sensor fusion stuff also. I would love to see this!

Though they talk about performance measures (timing results) for their algorithm, I'm not sure how indicative those are of general purpose results, as the size of their database grows with not only the region explored, but with the different views an object is seen from as well as possibly 3rd party motion pollution. Any thoughts on this stuff's scalability?