Can we dynamically change the hfield_size at runtime

[Suke0811]

Intro

Hi!

I am a PhD student at UCLA, I use MJX for my research on a humanoid robot.

My setup

MJX=3.3.1, WSL on Win11.

My question

We noticed that the in MJX, the mjx_model.hfield_size is numpy array and not jaxlib.xla_extension.ArrayImpl.

Hence, I am wondering if it is possible to change the hfield_size at runtime when batched with brax, i.e.

• related: #2406

Minimal model and/or code that explain my question

Code:

# scene from MJX Example
mj_model = mujoco.MjModel.from_xml_path('scene_hfield_mjx.xml')
mj_data = mujoco.MjData(mj_model)
mjx_model = mjx.put_model(mj_model)
# This is numpy
print(type(mjx_model.hfield_size))
# This is jax
print(type(mjx_model.hfield_data))

Confirmations

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

[Suke0811]

I came to a solution to this, and if you have the same issues, here is how you can deal with them.

---

hfield_size: (radius_x, radius_y, elevation_z, base_z) is not batchable, meaning all the parallel environments will have the same values.
However, the hfield_data, representing each node height normalized in [0, 1], is batchable. Hence, you can set the elevation_z to be the maximum height you want in the randomization. If you want to reduce the roughness i.e. 10% of elevation_z, then your hfield_data should be in the range of [0, 0.1].

This method changes the hfield contact, but the visualization will use the initial hfield_data values.