Offline reinforcement learning (RL) algorithms have shown promising results in domains where abundant pre-collected data is available . However, sharing data across all tasks in multi-task offline RL performs surprisinglypoorly in practice . To address this challenge, we develop a simple technique for data-sharing in multi task offline RL that routes data based on the improvementover the task-specific data . We call this approach conservative data sharing(CDS), and it can be applied with multiple single task offline offline RL methods . CDS achieves the best or comparable performance compared to prior offline multi task RL methods and previous data sharing approaches. On arange of challenging multi- task locomotion, navigation, and vision-basedrobotic manipulation problems, CDS achieved the best of the CDS has the most or similar performance compared with prior methods and other methods and approaches to previous data-share approaches, we say. CDS can also be applied to multi-Task locomotion and behavioral problems. It can be used

Author(s) : Tianhe Yu, Aviral Kumar, Yevgen Chebotar, Karol Hausman, Sergey Levine, Chelsea Finn

Links : PDF - Abstract

Code :

Keywords : data - task - offline - multi - sharing -

Leave a Reply

Your email address will not be published. Required fields are marked *