Updating & Testing 3CC-r Fatigue Model

[fl0fischer]

1. fixed bug in update formula of resting motor units (this might have caused fatigued motor units to not rest properly, and consequently MA, MR and MF not summing up to 1) -> if e.g. zero activations are used for a short period of time, fatigued units now rest, and can be activated shortly after again (see Figure below)
2. compute muscle force development and relaxation factors LD and LR based on the "effective time constants" internally used by MuJoCo (see https://mujoco.readthedocs.io/en/stable/modeling.html#muscles for details) -> peaks that previously occurred when the desired muscle activation changed rapidly are now smoothed out (see Figure below)

TODO:

• test fatigue model for models other than myoElbowPose1D6MRandom
• test whether including a smooth transition phase of length "tausmooth" as done by MuJoCo further improves the fatigue predictions
• check whether the second change introduces a significant deterioration in running time (?)

[fl0fischer]

If needed, I can also provide the modified testing script with which the comparison plots were created.

[fl0fischer]

@fl0fischer let's merge your branch into the myosuite/fix_fatigue and we can then iterate directly there

Sounds good.

[fl0fischer]

3. Modified clipping dynamics -> The muscle activation-deactivation drive C(t) is now clipped before the three motor unit compartments are updated. This ensures that the equation (MA + MF + MR = 1) is always satisfied.
   The resulting muscle activation trajectories are smooth now and do not exhibit the artefacts that previously occurred after a rapid change in control.