Question about Simpler_env Deployment

[yxy-2002]

Hello! Thank you for your team's excellent work on this project.

I am trying to run your models in the Simpler_env simulation, as I do not have access to a physical WidowX robot arm. I successfully followed your instructions for deploying the OpenVLA model in Simpler_env, and it works perfectly.

However, when I tried to use a similar approach to deploy the Emma-X model, I encountered an issue. The model outputs text in the correct format, but the actions seems incurrect. For example, the reasoning mentions "…closing gripper", but the output gripper action is still 1, which means gripper opening.(as Fig1&2 show) Also, the gripper positon sometimes could be incurrent, showed in Fig3.

I was wondering, does the Emma-X model need to be fine-tuned for the tasks in Simpler_env? Or are there some specific steps or configurations that I might have missed?

Thank you for your help!

[Akila-Ayanthi]

Hi @yxy-2002
did you figure out anything related to this?

[yxy-2002]

Hi @yxy-2002 did you figure out anything related to this?

no... i think maybe the model must be finetuned in the simulate env. But i didn't try it.