rain

> I now suspect this might be the result of bug fix 12 in 3.1.0. Could you please compare 3.0.1 and 3.1.0 to help us narrow it down? As per...

Thank you for linking those resources, very informative! I didn't think to look into the XML for options regarding this. Taking a closer look at the model trying to implement...

> I tried to rebuild the apps using updated libs from this comment: [fent/node-ytdl-core#1216 (comment)](https://github.com/fent/node-ytdl-core/issues/1216#issuecomment-1525777869) and its working 😄 I pushed an image to docker hub incorporating these fixes until...

I have made progress in identifying the underlying issue, but I am really not sure how to proceed. After getting MuJoCo building with debug symbols and getting C++/Python mixed debugging...

Update for anyone reading this issue in the future trying to migrate from MuJoCo 2: I did some more digging and realised that due to this `weldcnt` bug in MuJoCo...

Did some digging into this because it was really bothering me and turns out the behaviour seems somewhat expected / it's not really distrax' fault. I think the conclusion I...

I initially got it working and visually verified that the arm position looked like MuJoCo 2 using `torquescale="5"`, then before making the PR I tried lower / more conservative values...

With regards to validation, I did not run much before opening this pull request as I was requested by another maintainer to open it as soon as possible to run...

[I ran some hyper parameter tuning](https://wandb.ai/evangelos-ch/mw-weld-sweep/sweeps/3eq8loax) and it looks like with torquescale set to around 8.2 we can get an even closer match ``` Observation loss (MSE): 1.9481068874279174e-08 Observation loss...

> 1. There is no need to try to match the behavior of mujoco==2.3.7 and mujoco==3.0.0 on top of which I am not sure if fine-tuning for 1 environment would...