The future of robot-human collaboration

Humans are difficult to simulate. We are quick, complex thinkers, and our exact actions aren't predictable. So how can we sufficiently train robots to collaborate with us?