I wonder what the reason is to not go humanoid but super-human? Lack of training data?
For example, having 3 arms would help a lot of tasks. Or having fingers with twice the length of human fingers and 4 joints on each finger could enable them to switch a headlight bulb on a French car.
The third arm would put a load in one direction of the robot. You can no longer balance by swinging 1 arm back, you have to swing both of them, or involve a leg motion you might not want to because the third arm has a more limited range of motion.
Hands are just hard: no one is building good hands yet because the materials science and motors isn't there to do it (see the production Atlas's with the 3 finger grippers).
And of course training data: we have a wealth of examples of how to move a bipedal platform around, but there are no 4-armed humans so you'd be figuring out the balancing from scratch (but it also has the same problem: 4 arms in physics terms is still basically 2: if you lose your balance you'll need to involve both arms to correct it).
For example, having 3 arms would help a lot of tasks. Or having fingers with twice the length of human fingers and 4 joints on each finger could enable them to switch a headlight bulb on a French car.