Soft policy optimization using dual-track advantage estimator

Yubo Huang, Xuechun Wang, Luobao Zou +2

In reinforcement learning (RL), we always expect the agent to explore as many states as possible in the initial stage of training and exploit the explored information in the subseq…