Uni-RL: Unifying Online and Offline RL via Implicit Value Regularization

Haoran Xu*1, Liyuan Mao*2, Hui Jin*3, Weinan Zhang2, Xianyuan Zhan4, Amy Zhang1
1UT Austin 2Shanghai Jiao Tong University 2Nanyang Technological University 2Tsinghua University
* Equal contribution
Paper Code

TLDR: Can we design a scalable RL framework that unifies online, offline, and offline-online settings?