I agree that the conditions you listed are necessary for being a successful researcher. However, even if somebody were to meet those crtieria, it seems like they would still have little to no impact, since robotics capabilities research is such a popular and capacity-constrained field.

I.e. if I became a researcher at Deepmind / other labs, I would probably be taking a spot that another strong researcher would otherwise fill, limiting my marginal impact. Thus, my impact as a researcher would be close to zero, while at a startup, I could at least potentially have some impact.

Do you generally agree with this assessment?

It seems to me that it doesn't really work that way. You're not a cog, and robotics capabilities research is hugely multidisciplinary, so you should be able to synthesize unique insights of your own. There are always, always open problems and issues that nobody else is looking at, even in crowded fields.

Besides, as a noob researcher, you'd mostly be doing somebody else's grunt work, at least for a while. Learning the ropes. Then, if you survive, you'd be able to devote substantial time to projects of personal interest -- and those might be high-impact, or at least appear that way to you.