A deep neural adaptive PID controller would be a bit overkill for a simple robot arm, but for say a flexible-link robot arm it could prove useful. They can also work as part of the controller for systems governed by partial differential equations, like in fluid dynamics.
These datasets do not exist, you got that right.
I highly doubt there is much AI deep learning needed to keep a robot arms PIDs accurate. That seems like something a regular old algorithm can do.
A deep neural adaptive PID controller would be a bit overkill for a simple robot arm, but for say a flexible-link robot arm it could prove useful. They can also work as part of the controller for systems governed by partial differential equations, like in fluid dynamics.