mj_inverse computed qfrc_inverse components

[scakki]

Intro

Hi!

I am Shivayogi, a graduate student at Michigan Tech, I use MuJoCo for my research on Musculoskeletal system simulation.

My setup

Mujoco Version: 3.2.7

def get_qfrc(model, data, target_qpos):
    data_copy = deepcopy(data)
    data_copy.qacc = (((target_qpos - data.qpos) / model.opt.timestep) - data.qvel) / model.opt.timestep
    model.opt.disableflags += mujoco.mjtDisableBit.mjDSBL_CONSTRAINT
    mujoco.mj_inverse(model, data_copy)
    model.opt.disableflags -= mujoco.mjtDisableBit.mjDSBL_CONSTRAINT
    return data_copy.qfrc_inverse
    
# Run simulation loop
for idx in tqdm(range(traj.shape[0])):
    target_qpos = traj[idx, 1:]
    qfrc = get_qfrc(model, data, target_qpos)
    data.qfrc_applied = qfrc
    mujoco.mj_step(model, data)

My question

I'm working on a floating body musculoskeletal simulation using MuJoCo, and while the simulation runs well overall, I'm encountering an issue regarding force decomposition. According to the MuJoCo documentation, the net force is computed as:

  qfrc_inverse = qfrc_applied + Jᵀ*xfrc_applied + qfrc_actuator

Since mj_inverse returns this aggregated qfrc_inverse, is there a recommended way to extract the individual contributions—specifically, to separate the contact forces (transformed by the Jacobian) from the actuator/joint forces? I need to isolate the applied joint forces to accurately compute muscle activations.

Minimal model and/or code that explain my question

No response

Confirmations

• I searched the latest documentation thoroughly before posting.
• I searched previous Issues and Discussions, I am certain this has not been raised before.

[yuvaltassa]

If you set ctrl and call mj_fwdActuation it will compute qfrc_actuator, which you can then subtract from qfrc_inverse.

[scakki]

Thank you for the suggestion.
However, because I'm working with a musculoskeletal model, I don't have a computed ctrl signal for the muscle actuators at this stage, so I can't use mj_fwdActuation to generate qfrc_actuator. Currently, I'm directly applying qfrc_inverse to qfrc_applied for validation of the simulation. While the simulation appears to be functioning correctly, I notice that qfrc_applied shows large forces across the floating body's 6 DOF—which I suspect represents the contact forces. I'm uncertain how to isolate the forces acting solely on the actuated joints for further computation of muscle activations. Any guidance on extracting these specific forces would be greatly appreciated.

For reference, my workflow is similar to this tutorial.

[yuvaltassa]

Hmmm, in that case I don't think what you want is possible. Your diagnosis of the situation sounds correct.