The project is still in the academic research phase and I don't think they're quite as far as actually implementing it in a car in the wild, although they're moving ahead with it as fast as they can, so they might have actually done this recently. It won an award at the last International Conference on Computational Photography (ICCP) because it's just such a cool idea, and it seems like it might actually be feasible to some degree (even despite all the issues people have mentioned here and elsewhere). At ICCP, it was still in simulation and prototype-with-rain-machine-in-the-lab mode.
Btw I highly recommend watching the ICCP talk, available to watch from that webpage, or for download here: http://www.cs.cmu.edu/~ILIM/projects/IL/smartHeadlight/media...