<ul><li>A technique called Change Based Exploration Transfer (CBET) is adapted for world model algorithms like DreamerV3.</li><li>CBET shows potential in improving DreamerV3's returns in the sparse reward environment of Crafter, but reduces returns and leads to suboptimal policies in Minigrid.</li><li>Pre-training DreamerV3 with intrinsic rewards does not immediately maximize extrinsic rewards in Minigrid.</li><li>CBET has a positive impact on DreamerV3 in more complex environments like Crafter, but may be detrimental in environments like Minigrid.</li></ul>

World Model Agents with Change-Based Intrinsic Motivation

Discover more