nefc and ncollison overflow (MJX + WARP)

[itaouil]

Intro

Hi!

I am a PhD student at TUM, I use MuJoCo for my research on humanoid loco-manipulation.

I recently switched to using MJX with WARP backend. In my MJX model I set the following:

<size nconmax="2000" njmax="500"/>

and when I create the the mjx_data I do the following:

mjx_data = mjx.make_data(controller.task.mj_model, impl="warp", nconmax=2000, njmax=500)

however, it seems like I might be missing something else in terms of setup because I keep getting the following warnings at runtime:

ncollision overflow - please increase nconmax to 20
nefc overflow - please increase njmax to 137

and I cannot understand why.

I am using:

mujoco 3.3.7 mujoco-mjx 3.3.7 warp-lang 1.9.1
any help is appreciated. Thanks :).

My setup

mujoco 3.3.7 mujoco-mjx 3.3.7 warp-lang 1.9.1 Python 3.12.11

My question

No response

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:

minimal XML

<mujoco>
  <worldbody>
    <light pos="0 0 1"/>
    <geom type="sphere" size="1" rgba="1 0 0 1"/>
  </worldbody>
</mujoco>

Code:

import mujoco
import mediapy as media
model = mujoco.MjModel.from_xml_string(xml)
data = mujoco.MjData(model)

with mujoco.Renderer(model) as renderer:
  mujoco.mj_forward(model, data)
  renderer.update_scene(data)

  media.show_image(renderer.render())

Confirmations

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

[btaba]

<size nconmax="2000" njmax="500"/> are not used

The warning happens during make_data? Those warnings can be ignored, and were fixed here I believe

[itaouil]

Hi,

Thanks for the quick reply.

It happens during runtime actually....

so should I just ignore them and they are not going to affect the simulation rollouts? I am using mujoco 3.3.7, mujoco-mjx 3.3.7 and warp-lang 1.9.1.

[btaba]

You should not ignore them if they occur at runtime (not just during make_data time). They should be modified until no warnings occur.

Right now it's a bit confusing because nconmax in mjwarp is naconmax / nenv. See the updated warning here

But please increase njmax to 137 doesn't make sense to me since you are setting it to 500.

[itaouil]

yeah exactly.

that is what also I could not understand.

Because from what I understood the only place where I should set nconmax and njmax is when I call make_data. I dont define nworlds as I think when using mujoco mjx with warp is not required.

if you have any idea why something like this could happen let me know and I could investigate on my side.

[btaba]

Can you inspect what njmax is here

mujoco/mjx/mujoco/mjx/third_party/mujoco_warp/_src/forward.py

Line 224 in 1bfa8b6

wp.printf("nefc overflow - please increase njmax to %u\n", nefc)

It gets set here

mujoco/mjx/mujoco/mjx/warp/forward.py

Line 2993 in 1bfa8b6

_d.njmax = njmax

if you're calling step

[itaouil]

Sure. I call step in this function:

x = x.replace(ctrl=u) x = mjx.step(model, x) # step model + compute site positions

Could you tell me how to access the njmax variable in this case?

Also... when I downgrade to mujoco and mujoco-mjx 3.3.6 the issue disappears...... but it seems to run much slower for the same amount of worlds is this normal?

[yuvaltassa]

cc @thowell

[youwyu]

I saw the same issue with most Rough environments in mujoco-playground. I just manually changed https://github.com/google-deepmind/mujoco/blob/9f674b04a9491c1221fb21d19a4ca8de366bcec0/mjx/mujoco/mjx/_src/io.py#L970 to _make_data_warp(m, device, 1000000, njmax) then it works. So I guess there is something messed with the logic naconmax = nconmax if naconmax is None else naconmax?

[btaba]

nconmax is simply being renamed. Your issue may have been related to google-deepmind/mujoco_warp#955 @thowell ?

[btaba]

@itaouil the issue should be fixed on 3.4, let us know if it's still an issue