Correctly scaling actuators with mjSpec

[aaprasad]

Intro

Hi!

I am a phd student at mit using mujoco for my research on C.elegans neuromechanics

My setup

mujoco/mjx 3.4.0, python, linux ubuntu.

My question

I followed the mjSpec.ipynb example for scaling a mujoco model to downscale my model but for the behavior of each model diverges. It seems like the actuator strength on the scaled model is much higher relative to the mass than it is in the unscaled version even though I am scaling the actuator gain as mentioned in the notebook. Here is a video of the unscaled vs scaled model with random noise injection:

Screencast.from.2026-01-16.12-42-53.webm

Screencast.from.2026-01-16.12-43-55.webm

Are there other parameters I should be scaling such as the joint armature?

Minimal model and/or code that explain my question

If you encountered the issue in a complex model, please simplify it as much as possible (while still reproducing the issue).

Model:
I've taken the C. elegans biomechanical model from this repo. It's based on the n -link swimmer from gymnasium and contains 25 body components with corresponding meshes. The total length of the model is 1mm. I am trying to simulate a worm on agar so I'm using air as the medium. It is constrained to 2D movement in the xy plane with 24 joints hinge joints actuated by 95 muscles. Each joint is actuated by a set of 4 muscles organized into dorsal/ventral and left/right. Except for the last one which is missing its ventral left muscle. In order to get the model to work efficiently with mjx I removed mesh contact computation and used collision geoms instead. I also only kept contacts between the floor and the body (ie no inter body contact).

I've attached a zip file containing the arena and body model I'm using to create the scene via mjSpec. I've also posted the unscaled vs scaled models generated by mjSpec as well as the code used to generate these specs.

model_scaling.zip

Code:
Here is my scaling code to make it easier to see in case theres something obvious that pops out:

def dm_scale_spec(spec, scale):
    scaled_spec = spec.copy()

    # Traverse the kinematic tree, scaling all geoms
    def scale_bodies(parent, scale=1.0):
        body = parent.first_body()
        while body:
            if body.pos is not None:
                body.pos = body.pos * scale
            for geom in body.geoms:
                geom.fromto = geom.fromto * scale
                geom.size = geom.size * scale
                if geom.pos is not None:
                    geom.pos = geom.pos * scale
            for site in body.sites:
                if site.pos is not None:
                    site.pos = site.pos * scale
                if site.size is not None:
                    site.size = site.size * scale
            for joint in body.joints:
                if joint.pos is not None:
                    joint.pos = joint.pos * scale
            scale_bodies(body, scale)
            body = parent.next_body(body)

    # if scale_actuators:
    # # scale gear
    for mesh in scaled_spec.meshes:
        mesh.scale = mesh.scale * scale

    for actuator in scaled_spec.actuators:
        # scale the actuator gear by (scale ** 2),
        # this is because muscle force-generating capacity
        # scales with the cross-sectional area of the muscle
        actuator.gear = actuator.gear * scale * scale

    # scale the z-position for all keypoints
    for keypoint in scaled_spec.keys:
        qpos = keypoint.qpos
        qpos[2] = qpos[2] * scale
        keypoint.qpos = qpos
        keypoint.qpos[2] = keypoint.qpos[2] * scale

    scale_bodies(scaled_spec.worldbody, scale)
    return scaled_spec

Confirmations

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