I've been reading about the Georgia Tech program, which has interesting aspects. But I can't swallow the gimmickry. Robotics is a great context, but these are still toy problems. My argument is that even beginners can contribute to a large, complex project. I've had freshmen and sophomores make significant contributions to research projects. Granted, those are some of the best students at Penn, but I'm convinced that with a careful structure, it is possible to engage many more into meaningful projects.
I've been in many discussions, panels on the drop in computer science enrollments. There are lots of theories about it, and preciously little empirical evidence. Some changes in freshman classes help. But most of the factors in the drop are likely to be beyond our control.
Instead of trying this and that in the hopes that it might make a difference, we should instead think about what an excellent computer science education should be like. The current models, even in their best manifestations, wring out much the main joy of the field, which is to muster deep thought and computational processes to solve real problems.