Electronic – Limiting torque on a DC motor

you could recreate that, but from a quick look at the design, it has several flaws. The fishing line used will be slightly elastic, limiting how hard a contact it can simulate. It only drives in one direction, so whilst it can simulate a grasp of a small object it cannot simulate interactions that force the hand closed. Also it is not grounded, so cannot simulate the mass of the object you are holding. Also you will have to have some constant torque on the cables to make them retract, even in free space.

You cannot use the type of motor you suggest for one simple reason. The device is by its very nature backdrivable. What this means is that when you are not contacting something the cable is pulled, and the motor spins. This means that you need the motor to have very low friction/inertia for your fingers to feel like they are in free space.

The best option is a coreless dc motor, as this has the lowest inertia. Using such a system when the user pulls producing more torque then the motor, the motor just spins backwards, with no damage to anything. You will probably need to only drive motors at continuous current, as in haptics you tend to drive actuators at stall for prolonged periods.

You did not mention what min and max speed is required. I suppose you don't want to make any compromise in any parameter ;-)

With stepper and microstepper driver you can achieve better than 0.15 degrees resolution. But what is precision ? Probably wrong because there is motor and microstepping non-linearity and torque is low, dynamic range is low and extra problems come at very low speed.

You can arrange it as dual loop control with load encoder at outer loop. It should improve precision. Let say 13bit load encoder and 32 or 64 microstepping with 400 step motor might satisfy 0.15 precision.

Gearbox helps with torque and resolution but you also need dual loop control and no backlash if you want change direction. There is extra problem if there is a spring in gearbox. If yes then there is a force at output you need high frequency control loop. It might be even impossible to control system when you must consider this force. So expensive gearbox or back to direct drive.

You can substitute stepper with brushless 3 phase motor and encoder with or without gearbox and one or two encoders. Dynamic range and price go up. Torque ??? For high speed you need high bandwidth encoder/decoder. And it's very difficult to debug such a setup in real time as you almost cannot use breakpoints.

Generally when using encoders they must satisfy your precision needs (and not resolution only).

EDIT (as commenting is forbidden): AS5045/8 encoder: I think you need consider mainly non-linearity INL parameters in range of degrees which affects accuracy (worser than 0.15 deg). Also propagation delay in range of 100 us limits speed, 1RPS = 1/4096 = 244us per position tick.