About this document

GO-DICE: Goal-Conditioned Offline Imitation Learning by Yücel Beyazıt is a document available to read on EtoBox.

GO-DICE is a novel offline imitation learning technique designed to improve learning of long-horizon tasks by utilizing goal conditioning and hierarchical policy learning from both expert and imperfect demonstrations. It segments tasks into sub-tasks to facilitate reasoning and minimizes retraining when task goals change, outperforming existing methods in challenging robotic tasks. The approach leverages auxiliary inputs, such as task segment annotations, to enhance learning performance in offline scenarios

Author

Yücel Beyazıt

Language

EN