Here is the 3rd example I've seen, (and I'm sure more to come) of dynamic Haar training and Bag Of Words association. It seems the little bot is not being sold right now, but I see the drool from the sidelines
Full link here - http://thecorpora.com/blog/
I'm pretty close to getting dynamic Haar training in MRL... although I'll be using the Kinect or interactive selection vs stereo cameras to isolate the object for association.
They definitely have the right idea about cloud processing too... MRL does cloud processing to make it's sexy "Audrey" voice.
QBo really looks like a very cool project.
https://www.youtube.com/watch?v=T4z11lIKnDU