I mean the world is usually (and should be) designed for people in wheelchairs, too. The only thing that comes to mind that humans frequently interface with, which benefits from human bipedalism, is stairs. Stairs aren't something that's impossible to design around with a wheeled vehicle, either, and I don't think the efficiency tradeoff is one that turns out in favor of legs, at least for robots.

The only thing I can really conceive of as being a use case for a humanoid robot is if you decided to deploy it in a totally unbuilt environment with little infrastructure, which is pretty counterintuitive considering the energy density of current battery technology. I think that would probably only work as an idea if you had like, a nuclear battery, for a human scale robot, and we have tank treads for everything else. I think maybe the prevalence of legs in robotics either stems from the fact that people think it's cool, which is important for funding, it stems from some amount of funding coming from military or pseudo-military applications like "disaster relief" where an ability to operate in diverse environments is seen as a plus, and it stems from people banking on denser and denser battery technology and maybe lighter weights material science.

I think it also stems from a kind of all-encompassing ideal to create a totally self-sufficient worker-slave that is both unconscious but is also totally adaptable to the environment and can operate with minimal inputs, compared to a person. There's like, some conception that you can make a robot which can act in an environment in the same basic way as a person, but then you also can't make any robots that are designed for any specialized tasks, and which might do those specialized tasks in a much, much more efficient way than a humanoid robot would be able to do them. Also somehow this ability to do tasks similar to how a person might do them would really be like, something that you have to confine to a human sized body, rather than just using it to sort of further automate the managerial class. There's some idea that this is more efficient to scale, rather than being more efficient at scale, or that somehow once you make a humanoid robot you will have cracked the code somehow, and everything will just be post-scarcity, because you can make the robots mine the lithium to make more robots.

I don't think I have to tell you that all of these are kind of naive, as viewpoints, or, are intrinsically viewpoints that kind of discount the amount of interpretive labor that has to be performed by people in order for the system to work, the variety of said labor as it exists, or the amount of effort involved that you would really have to do in order to automate that away. I think, to put it more bluntly, if you tasked a robot with automating a kind of, big, general batch of tasks like humans might perform, it would probably make a specialized set of robots which can complete all the tasks as suited to the task each robot has to do, rather than creating a big general human shaped robot to do them all.