Eval Policy Mismatch (EEF Space) for Pretrained Joint Model

[D222097]

Hi, I'm encountering an issue while evaluating a pretrained policy and have two key questions related to action/state space compatibility:

Eval Policy Action Space Mismatch:
I’m training a pretrained model on a large joint-space dataset but want to evaluate this policy on the Libero environment. However, I’ve noticed the Libero environment’s execution action space is in EEF space, which causes a mismatch with my pretrained model.
Is there a way to resolve this mismatch and successfully evaluate my joint-space pretrained model on Libero?

State Space Dimension Clarification:
I’m also confused about the state space dimension in the Libero environment: the state dim is reported as 8. Could you confirm if this 8-dimensional state represents:
Joint space or EEF space (e.g., EEF + gripper state + gripper state, etc.)?

Any help or suggestions would be highly appreciated! Thanks for your help!