Lukas Hermann

It renders in headless mode by default. which error do you get?

Are you sure the nvidia-drivers are correctly installed? What's your output for `nvidia-smi` ? The output of `list_egl_options.py` should list the Nvidia card.

1. Yes, we run a validation step after every training step. We separated the data into different folders to make sure the sliding window approach never samples a frame in...

Yes, you can run the CALVIN evaluation as described in the README [here](https://github.com/mees/calvin?tab=readme-ov-file#muscle-evaluation-the-calvin-challenge) and if you get a performance that beats our baselines (incl. HULC), then contact [Oier](https://www.oiermees.com/) to add...

Yes, there is [calvin_env](https://github.com/mees/calvin_env), a simulated environment running on the pybullet physics engine. We use it during the evaluation, have a look at the [code](https://github.com/mees/calvin/blob/main/calvin_models/calvin_agent/evaluation/evaluate_policy.py).

You can run `calvin_env/calvin_env/envs/play_table_env.py` and it will open the pybullet debug viewer where you can see the environment. To interact with the environment, you would have to provide the robot...

Does [this](https://github.com/mees/calvin/tree/main/dataset#state-observation) answer your question?

@Kira0096 did you mean recording new play data with manual teleoperation or just defining and detecting new tasks in the existing environment interactions?

The language dataset is 1% of the vision dataset that has been labeled with language instructions. In Multi-Context Imitation Learning, the agent is trained with different goal modalities, either a...

1. This doesn't seem to be caused by calvin. Did you try running wandb in dryrun? Setting the environment variable `WANDB_MODE="dryrun"` should turn off the sync. Alternatively you can also...