Discourse around physical AI has mostly focused on answering the questions: what can humans teach robots?

Without a doubt, it's a worthy and pragmatic question but I think the converse is marginally more interesting : What can robots teach humans?

on ergonomics, a quick idea

Many product and ergonomic decisions on consumer hardware are done in a psueo-scientific manner. Even with subjective accounts from focus groups or with a/b testing or even with designing to best practices set by a regulating body, quantifying ergonomics and a customer's experience is just a very difficult thing to quantify.

how easy is it to get into a car?

Most people get into a vehicle a similar way.

What if we used a humanoid robot to answer : how easy is it to get into a car?

Say you train a VLA to ingress and egress a vehicle. Not sure what exactly you would train on but I'm sure there's enough clean data of this. Then you could use some advance force sensing or read motor effort to determine difficulty.

This is just a half baked idea, but thought I would write it down so I didn't forget it.