R2I introduces a new method, Recall to Imagine (R2I), integrating state space models (SSMs) into world models of model-based reinforcement learning agents. This integration aims to improve temporal coherence, long-term memory, and credit assignment. Through various tasks, R2I establishes a new state-of-the-art for challenging memory and credit assignment RL tasks. It showcases superhuman performance in the complex Memory Maze domain while maintaining comparable performance in classic RL tasks like Atari and DMC. R2I is faster than the state-of-the-art MBRL method, DreamerV3, resulting in faster convergence time.
翻译成其他语言
从原文生成
arxiv.org
更深入的查询