The problem with search and rescue is determining what needs to be rescued. How does an autonomous robot know what a human is? That is the million dollar question. Facial recognition has tons of money thrown into by private companies and governments, but if the face is not viewable at the very specific angle it is no good. You could do form recognition, but in the snow or in the water the form is going to be atleast partially covered. Plus a person can not only have a thousand different body styles, but also will be positioned differently. Heat variances from the terrain could work, but there you still need a way for the robot to know the difference between a person in need of rescue, a person who just happens to be in the robots search vector, and people actually searching alongside the robot.
I'm hoping for a robot that is relatively easily duplicatible by another hobbiest, not a multimillion dollar government contract.