AttributeError: 'State' object has no attribute 'pos' in mjx/training_apg.ipynb

[MacQWER]

Intro

Hi!

Recently, I have been using MuJoCo for my research on First-order Policy gradients (FoPG).

My setup

Colab and Ubuntu 20.04
mujoco 3.3.0
jax 0.5.3
brax 0.12.1

What's happening? What did you expect?

I am encountering an AttributeError: 'State' object has no attribute 'pos' when running the train_apg.ipynb example. The error occurs specifically when calling the train_fn function as follows:
make_inference_fn, params, _= train_fn(environment=env,
progress_fn=progress,
eval_env=eval_env)
And the screenshot of the error is :

I have confirmed that the same code runs without issues in my old environment with mujoco 3.3.0, jax 0.5.3, brax 0.12.1. However, in my new environment with MuJoCo MJX 3.3.5, the problem consistently appears. This suggests that the training_apg.ipynb example may be using an outdated API that is no longer compatible with the latest MuJoCo MJX version.

Steps for reproduction

Just run training_apg.ipynb, with the new version of mjx and brax, this problem can be reproduced.

Minimal model for reproduction

Code required for reproduction

Confirmations

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

[Vaidik1199]

Chat Gpt said:

The error comes from an API change in recent MuJoCo MJX. In older versions, the State object exposed a .pos field, but this is no longer available in 3.3.5 and later. Instead you should use the new attributes such as qpos for generalized positions.

That means the example train_apg.ipynb is written against an older MJX API and needs to be updated. If you want to run it as-is, you can stick with the older MuJoCo 3.3.0 where the pos field still exists. If you want to use the new MuJoCo MJX, replace calls to state.pos with the correct field (state.qpos or the matching attribute used in your environment).

This is not a bug in MuJoCo itself but rather an incompatibility between the example notebook and the updated API.