Their benefit is they can operate in any space designed for a human without custom modification. No wires to get in the way with an assistant robot that interacts with the existing built world.

Ish. Millions of years of evolution and other than snakes and their ilk, land-faring animals have legs. So bipedal with wheels on the feet is maybe optimal. The optimal number of arms I think is three. No "head" is strictly necessary. The interesting question is how do you train a three armed robot when there are few three armed people?

> almost certainly suboptimal.

I agree from an engineering and cost perspective that matching the embodiment, degrees of freedom and performance to the use case tends to be optimal. But entertainment, fun and inspiration can also be economically valuable use cases - even if less common.

That said, there are use cases which require navigating environments built for humans and the adaptive flexibility to perform unpredictable tasks as a human proxy. And those robots will probably be broadly humanoid shaped.

How does R2D2 climb stebs?

We might be at the beginning of a golden age for inventors. Robots of all shapes and sizes, with highly variable price points and specialization. Still, the first challenge for R2D2 is probably stairs.

The big benefit of humanoids is that they can be trained from captured human actions (egocentric datasets made by humans).

> The desire for humanoid robots is purely aesthetic and almost certainly suboptimal

Based on what? Do you think humans evolved into what we are based purely on aesthetic? Of course it may be suboptimal, but given enough time and ingenuity, so could literally everything be. Not sure that should stop anyone.

> Humanoids seem like a real dead end to me.

We built the world for us, not for the most sensible robot.

So for generalized solutions, a good biped will have some advantages.