Running on AWS EC2

[AvivCommuna]

Intro

Hi!

I'm a professor at Technion.

Use Mujoco for robot learning.

My setup

Mujoco 3.2.7
Python
EC2 instance type: g4dn.2xlarge
AMI: Deep Learning OSS Nvidia Driver AMI GPU PyTorch 2.5.1 (Ubuntu 22.04) 20250216

My question

I am trying to run Mujoco on an AWS EC2 instance with GPU and have issues with headless rendering.
I have tried both EGL and OSMesa but get the same error when running mj.MjrContext:
mujoco.FatalError: gladLoadGL error

Would really be grateful if anyone managed to run mujoco headless on EC2 and can share how.

Minimal model and/or code that explain my question

Here's a failure case with Osmesa. I get the same error if I change osmesa to egl.

import os
os.environ["MUJOCO_GL"] = "osmesa" # Use OSMesa for headless rendering
os.environ["PYOPENGL_PLATFORM"] = "osmesa" # Force PyOpenGL to use OSMesa

import mujoco as mj

minimal_xml = """

"""

model = mj.MjModel.from_xml_string(minimal_xml)

context = mj.MjrContext(model, mj.mjtFontScale.mjFONTSCALE_100.value)

Confirmations

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

[AvivCommuna]

This appears like an issue with using MjrContext on a headless machine, and nothing specific to EC2.
I get the same problem on the official Mujoco tutorial Colab
(https://colab.research.google.com/github/google-deepmind/mujoco/blob/main/python/tutorial.ipynb)
if I add, e.g.,
mujoco.MjrContext(model, mujoco.mjtFontScale.mjFONTSCALE_100.value)
after the code for creating the model.

Since headless rendering works on the tutorial colab, and the mujoco.Renderer wrapper successfully creates a context internally, I was able to get my code to work by copying the code from mujoco.Renderer:

from mujoco import _enums
from mujoco import _render
from mujoco import gl_context
_gl_context = None # type: ignore
if gl_context.GLContext is not None:
_gl_context = gl_context.GLContext(320, 240)
if _gl_context:
_gl_context.make_current()
context = _render.MjrContext(
model, _enums.mjtFontScale.mjFONTSCALE_150.value
)

This works on the colab.