Model-based Offline Policy Optimization with Adversarial Network (MOAN)

Offline RL framework called MOAN, which introduces a two-player game to improve the generalization capability of the transition model and mitigate the negative effects of potentially problematic rollouts during offline reinforcement learning.