Problem

When self.actor_params_generator is not None, we call the function get_attr_val_from_sample, but the definition of this function was in rlgpu which is no longer available: from rlgpu.tasks.base.base_task import get_attr_val_from_sample. This results in an error when trying to run the code with customized domain randomization.

Solution

To fix this issue, we need to provide a replacement for the get_attr_val_from_sample function within the codebase.

I think the code does not support the currently available IsaacGym package, and the preview_2 version is not available any more. This breaks the PPO training.

Is there plan to update for the newest IsaacGym? Thanks!

I have followed all instructions for installation. Also exported the precise variable for isaacgym. However bayes-sim-ig is unable to source rlgpu. The error points to line 36 in bayes_sim_main.py file where rlgpu is referred. Is there any variable that can be set to point to rlgpu?