Behavior-Invariant Task Representation Learning with Transformer-based World Models for Offline Meta-Reinforcement Learning | ArxivCSExplorer