Feedback for getting started

[vyeevani]

There's a few hangups that I had while getting started training my own policy with this dataset that I wanted to capture as feedback to the maintainers.

1. Given that the training scripts are expecting tfrecords, I wonder if the tfrecord + numpy formatted datasets could not be included out of the box
2. The size of the dataset means that it's fairly challenging to get up and running with experiments quickly. It took me a few hours just to download the dataset and unpack
3. Evaluation and debugging black box policies has historically been quite challenging, but simulation with a standard set of environments with a standardized set of camera positions and tasks would be a place to start in addition to real world testing. Maybe this is a request best made to interbotix though

[vyeevani]

I filed: Interbotix/interbotix_ros_manipulators#120