I think this is a bit too ambitious. Let's first have 1 robot reliably do things. And only then generalize.