The TLDR here is that if you want agents that can survive autonomously (as in - without help from a human) - your best bet is to rewire their brain via reinforcement learning to be good at exactly that. And no one is working on this (so we will).