RMBench evaluation recipe for the released visual-anchors-only checkpoint

[zoeyliu1999]

Hi authors, thank you for releasing EventVLA and its checkpoints!

The README marks “EventVLA training and evaluation code in simulation (RMBench and RoboTwin-MeM)” as released, and states:

The RMBench checkpoint should be used with the corresponding RMBench environment and evaluation configuration.

I am trying to evaluate the released checkpoint

• RMBench/final_model/pytorch_model.pt

and reproduce the visual-anchors-only result reported for RMBench in Table 8 (67.8% average success rate).

However, I could not find

• an examples/RMBench evaluation entry
• or the corresponding evaluation configuration in the current repository.

Could you point me to, or release, the exact RMBench evaluation recipe/config used with this checkpoint to obtain the reported result?

Thank you!