I'm planning the design of a wrist for a humanoid robot. I would like to choose a design that is sturdy while allowing for dexterity comparable to a human wrist.
One option that was presented to me was to use a Stewart platform. This setup appears to correctly recreate all possible movements of the human hand. My immediate concern is that this platform will use a total of six actuators which will require additional power and computational requirements. I don't want to commit to this design until I am certain that there isn't a better alternative.
Is a Stewart platform a good choice for replicating the dexterousness of the human wrist? If not, what is a better solution?
The "best" robot wrist in terms of human analog is probably the omni-wrist by Mark Rosheim. It has a large range of motion, and does not have singularities or gimbal lock that plague other more conventional wrists. However, it is fairly complicated mechanically, and thoroughly patented i believe.
Bear in mind a Stewart platform has six degrees of freedom - yes it does reproduce the main motions of the human wrist, but it adds extra ones such as the ability to change its overall length. There is (I would guess) additional complexity in building one - finding suitable compact actuators and linkages, additional work to control its path, perhaps robustness.
If your objective is just to get a simple arm working, I would look for a more conventional three axis design. However the Stewart platform idea does add extra features - those extra degrees of freedom - and might be seen as a more novel concept.