State Advantage Weighting for Offline RL

Jiafei Lyu, aicheng Gong, Le Wan, Zongqing Lu, Xiu Li

TL;DR: We investigate QSS learning for offline RL, where we leverage state advantage weighting for update.

Abstract: We present \textit{state advantage weighting} for offline reinforcement learning (RL). In contrast to action advantage $A(s,a)$ that we commonly adopt in QSA learning, we leverage state advantage $A(s,s^\prime)$ and QSS learning for offline RL, hence decoupling the action from values. We expect the agent can get to the high-reward state and the action is determined by how the agent can get to that corresponding state. Experiments on D4RL datasets show that our proposed method can achieve remarkable performance against the common baselines. Furthermore, our method shows good generalization capability when transferring from offline to online.

Community Implementations: [ 3 code implementations](https://www.catalyzex.com/paper/arxiv:2210.04251/code)