Advantage-Weighted Regression
Imitating only good transitions based on how good the actions are, with weighting each transition depending how good the action is
Can show that advantage-weighted objective approximates KL-constrained objective.
AWR
Creator
Creator
Seonglae ChoCreated
Created
2024 May 1 2:24Editor
Editor
Seonglae ChoEdited
Edited
2024 May 1 2:33Refs