questions about load dataset for localhost, make own policy and use headless eval mode

[zhouzhq2021]

Hello, I'm trying to download a data set on hugging face to the local and then call this data set from the local. For example, 'aloha_sim_insertion_scripted_image' , its format is many 'episode_000000.parquet' files . Then how to load this format by LeRobotDataset() func or other ways?

Second, I want to create my own policy. After I parse the code framework, I think I may need to create my policy code file by mimicking the following files:

• lerobot/common/policies/act/configuration_act.py
• lerobot/common/policies/act/modeling_act.py
  However, I am having some difficulties in making my own policy now, and I want to create a new policy to implement my idea, which is to introduce the concept of comparative learning. That is to say, the policy enables the agent to learn the correct samples and stay away from the wrong samples. I would like to ask you what should be modified to complete this idea.

I really need examples of this, and it would be very helpful if you could give me detailed advice!

Finally, my server is headless, which means that when evaluating a policy, there is no way to call mujujo to view the evaluation, so can our code framework support headless mode and save the evaluation video?

As a new researcher in this field, it would be great if I could further communicate with you about the above issues. Thank you very much!

Best wishes : )

[zhouzhq2021]

I use xvfb-run python lerobot/scripts/train.py command solved the last problem