Online Reinforcement Learning With World Model Adaptation

* ⏫ 02/2023: [Grounding Large Language Models in Interactive Environments with Online RL (GLAM)](https://arxiv.org/abs/2302.02662)

Unique: DreamerV3 supports online RL through continuous world model updates on a mixture of old and new data, enabling adaptation to environment changes. The design uses a replay buffer to balance stability (learning from diverse data) with adaptation (incorporating new information).

vs others: Enables continuous adaptation to environment changes while maintaining stability through replay buffer-based training, outperforming naive online learning approaches that update only on recent data.