Comment by cpgxiii

If the current->torque behavior of motors with gearboxes could be generally well characterized, the entire distinct market sector of "cobots" would not exist because every industrial robotics vendor would have long had good torque modelling on their actuators.

As it stands, Universal Robots (and likely their clones) do use current->torque characterization for their actuators (which, amusingly, is then stored on a robot-specific USB drive or SD card), and their torque sensing is shit. Shit enough that for any useful force/torque application you still need a separate force/torque sensor. Schunk, for some of their electric parallel grippers with "force" feedback, only characterizes them at a single velocity and there is significant error in the force estimate at any other speed. Good current->torque characterization of a complete actuator is so difficult that approximately no vendors in the automation space are willing to do it.