Hi, thank you for the great work on this repo!
In the async_cable_route_drq task, I noticed that the record_demo.py script requires a pre-trained classifier checkpoint_path to initialize the BinaryRewardClassifierWrapper. However, I thought this script was designed to record both positive and negative demonstrations to train the classifier from scratch.
Am I missing something?
Without a pre-trained classifier, is there a way to record demonstrations manually (e.g., by labeling episodes as positive/negative)?
How do I handle the task when there is no classifier to automatically segment episodes and labeling the frames, as was done in peg_insert?
Looking forward to your insights.
Thanks!
Hi, thank you for the great work on this repo!
In the
async_cable_route_drqtask, I noticed that therecord_demo.pyscript requires a pre-trained classifiercheckpoint_pathto initialize theBinaryRewardClassifierWrapper. However, I thought this script was designed to record both positive and negative demonstrations to train the classifier from scratch.Am I missing something?
Without a pre-trained classifier, is there a way to record demonstrations manually (e.g., by labeling episodes as positive/negative)?
How do I handle the task when there is no classifier to automatically segment episodes and labeling the frames, as was done in
peg_insert?Looking forward to your insights.
Thanks!