K-Scale has an updated fork of [[Humanoid Gym]] available at this [https://github.com/kscalelabs/humanoid-gym GitHub repository].

One of the significant changes involved modifying how the framework handles the initialization of simulation data. Previously, the framework used a fixed dimension for reshaping tensors, which limited its flexibility in handling different numbers of bodies in the simulation. The recent update adjusted this process to allow for any number of bodies, thereby improving the framework's ability to manage various simulation scenarios more effectively.

Another important enhancement was the optimization of graphics device handling, especially in headless mode where rendering is not required. The update introduced a conditional setting that disables the graphics device when the simulation runs in headless mode. This change helps improve performance by avoiding unnecessary rendering operations, making the framework more efficient for large-scale simulations where visual output is not needed.