Config and Checkpoint Issue

[annahsu041]

Hi. Thanks for your work. I'm encountering two different errors when trying to evaluate a GR00T-N1 policy using in the IsaacLabEvalTasks repo. Below are the details for each case.

case1: When I followed the instructions in and ran the following command:
python scripts/evaluate_gn1.py \ --num_feedback_actions 16 \ --num_envs 10 \ --task_name nutpouring \ --eval_file_path $EVAL_RESULTS_FNAME \ --model_path $CKPTS_PATH \ --rollout_length 30 \ --seed 10 \ --max_num_rollouts 100
It showed error says KeyError: 'gr1_arms_only'.

case2: shape mismatch during modality transformation
I swithced to different config like below:
python scripts/evaluate_gn1.py \ --task Isaac-NutPour-GR1T2-ClosedLoop-v0 \ --data_config **fourier_gr1_arms_only** \ --model_path /home/ee904/IsaacLabEvalTasks/GR00T-N1-2B-tuned-Nut-Pouring-task \ --num_envs 10
This triggered a warning that some weights of were not initialized from the model checkpoint, and then failed with the following error:
ValueError: Error applying transform 10 to data: index 1 is out of bounds for axis 0 with size 1

1. What is the correct value for when using the task?
2. How can I resolve the transform error in Case 2? Is there a specific input format or preprocessing step required?
Thanks in advance for your help!