I’m very interested in autonomous vehicles and I’m always wondering what the biggest challenges are. I’m not worried about public acceptance or anything like that. I’m more interested about the technical challenges of the software stack: perception, prediction and planning.
I’ve come to the conclusion that prediction is the big problem. If perception and prediction were solved, planning would not be a serious show stopper for self-driving cars.
From a perception point of view I always wonder how localisation is done based on radars and cameras (Tesla approach) assuming GNSS outage and no HD maps. That’s another potential weak link in the chain.
Let me know what you think.