VLABench evaluation with GR00T policy

[ryuji-oi-artic]

Hi, thank you for your great work!

I am trying to fine-tune the GR00T N1.6 policy on VLABench datasets using LeRobot library.
While exploring the repository, I found VLABench/evaluation/model/policy/gr00t.py, but I couldn't find documentation on how to use this specific script.

Could you clarify the intended usage? Specifically:

1. Is this script designed to act as a client that communicates with an official NVIDIA GR00T repository acting as the policy server?
2. Alternatively, is it compatible with LeRobot as a policy server?

Any guidance or examples on how to set up the evaluation pipeline would be greatly appreciated. Thank you!