DROP: Dynamics-aware representation learning for offline reinforcement learning

Jun Zheng

doi:10.1016/j.asoc.2025.114525

DROP: Dynamics-aware representation learning for offline reinforcement learning

2025-12-23

0

PRE

AI

DOI：10.1016/j.asoc.2025.114525

Original

Original request for help

Save

Abstract

En 中文

• DROP learns dynamics-aware joint state–action representations for offline RL. • Jointly predicts rewards and next-state embeddings to capture transition structures. • Incorporates behavior cloning regularization directly into Q-function learning.

DROP: Dynamics-aware representation learning for offline reinforcement learning

Abstract

Journal

Organization