[Bug Report] MuJoCo/Walker2d left foot has different friction than right foot

[Kallinteris-Andreas]

Describe the bug

Right foot has a friction coeff of 0.9
https://github.com/Farama-Foundation/Gymnasium/blob/main/gymnasium/envs/mujoco/assets/walker2d.xml#L25
left foot has a friction coeff of 1.9
https://github.com/Farama-Foundation/Gymnasium/blob/main/gymnasium/envs/mujoco/assets/walker2d.xml#L38

This issue was previously reported, but not addressed.
openai/gym#1068
openai/gym#1646

from what I can tell, this is most likely a typo
both were supposed to be 1.9 (or 2.0)

Since Walker2d and Hopper will be getting new models in v5 (cause google-deepmind/mujoco#833)
should this be fixed there?

Code example

No response

System info

No response

Additional context

No response

Checklist

• I have checked that there is no similar issue in the repo

[pseudo-rnd-thoughts]

@Kallinteris-Andreas Yes that is crazy it wasn't fixed at the time for v4.

This should affect the performance of older models but could you test training performance with 0.9 or 1.9 to know if it matters which value we use?

[Kallinteris-Andreas]

The coeff supposed to be 1.9 or 2.0, 0.9 does not make sense,

I will do a minimal test soon(-ish)

[Kallinteris-Andreas]

SB3 seed=0 (v4 modes were used a not v5)

i can verify that the friction coeff does matter significantly impact the model unlike what was suggested in the previous issues.

I will go with 2.0 since this is the value in Hopper, and Walker2D is supposed to be "double Hopper".

[pseudo-rnd-thoughts]

What value is this with?

[Kallinteris-Andreas]

friction=2 it says so in the legend
v4 is what is shipped in Walker2D-v4

[Kallinteris-Andreas]

Setting feet friction to 2.0
results in a model learning this policy:
https://raw.githubusercontent.com/Kallinteris-Andreas/gym-mjc-v5-model-validation/main/walker_feet_friction%3D2.gif
Essentially a 2 legged hopper not a 'walker'

policy was trained with SB3/TD3 2mil steps

I will try settings feet friction to 1.0 to see if we a get a model that can efficiency walk.

Note: from the research [Walker2d-v1](https://github.com/openai/roboschool/blob/master/roboschool/mujoco_assets/walker2d.xml

If any of you know where I could find any development documention as to why the models were designed the way they are, please let me know

[Kallinteris-Andreas]

I have set feet friction to 1.9, it appears to be walking (note: 2.0 would likely also be able to walk, but on that it may have developed that policy)