MocoTrack noisy actuators controls

[Mohammadreza Rezaie]

Hi, I'm aiming to run a torque-driven marker + contact tracking simulation with weak [1N(m)] residual and strong [200N(m)] reserve actuators. The simulations converged successfully, and the markers and GRF were tracked well with the max abs residual actuation of 0.01 N(m). However, the actuators controls are not smoothed at all. I tested increasing/decreasing the weight of the control-effort goal (1 to 1e-4), constraint and convergence tolerances (1e-2 to 1e-6), as well as almost all available options in the solver, such as implicit/explicit multibody dynamics, backward/forward finite difference, but couldn't see significant improvement there. Here is my best output:

graph_grf.png (57.26 KiB) Viewed 619 times

graph_joint_angle.png (98.11 KiB) Viewed 619 times

graph_joint_moment.png (107.39 KiB) Viewed 619 times

I was wondering if you could guide me in this regard. Is there any option to minimize the derivatives of the actuators controls?

Thank you in advance.

[Nicholas Bianco]

Hi Mohammadreza,

The actuator controls don't seem particularly "noisy", but there are certainly spikes at a some time points. Since you are running the problem with weak residuals and tightly tracking markers and GRFs, the model will need to adjust the controls (compared to ID) in order to maintain dynamic consistency.

What do the residual forces from inverse dynamics look like? I would only expect the controls to match closely if the ID residuals were already close to zero.

-Nick

[Mohammadreza Rezaie]

Hi Nick, thanks a lot for your response.

The residual forces/moments in ID are huge (<380N, <50Nm). However, I don't want the simulated joints moments match the ID output. I just want actuators control to be smooth and natural (without spikes) given that the kinematics and GRF look good.

I just found that the COPs were not tracked well, which may propagate error to the proximal joints. Do you have any idea how to improve it? Do I need tighter convergence tolerance?

Screenshot 2024-08-16 031954.png (33.86 KiB) Viewed 484 times

Thanks again for your help.

[Nicholas Bianco]

Hi Mohammadreza,

If you have large residuals in ID, then it is unlikely that you will be able to achieve fully smooth controls by tracking the markers and GRFs tightly. Since you are enforcing the residuals to be small, the dynamic inconsistencies have to be made up somewhere.

If you want smoother controls, you will likely need to decrease the tracking weights on your reference dat. Usually we trust GRFs signals more and allow kinematics to deviate, but you will probably need to allow at least slight deviations from the GRFs to get the controls you want.

Best,
Nick

[Mohammadreza Rezaie]

If you have large residuals in ID, then it is unlikely that you will be able to achieve fully smooth controls by tracking the markers and GRFs tightly. Since you are enforcing the residuals to be small, the dynamic inconsistencies have to be made up somewhere.

This is indeed a nice justification. I will explore more.
Thanks for your help.