World Model Agents with Change-Based Intrinsic Motivation

Jeremias Ferrao, Rafael Fernandes Cunha · 2025

Read World Model Agents with Change-Based Intrinsic Motivation by Jeremias Ferrao, Rafael Fernandes Cunha (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-21047.txt.

Description

Game-like worlds with rare rewards are hard for learning agents, since good moves may pay off only after long chains. Jeremias Ferrao and Rafael Cunha adapt a curiosity method called CBET for world-model agents in a 2025 paper. The method rewards an agent for change it has not seen, and it can carry prior learning into a new task.

The pair tested the DreamerV3 agent, with and without the added drive, in two sparse-reward worlds. In the richer world, Crafter, the drive raised scores. In the simpler Minigrid world, scores fell, and prior training did not lead to a good final policy. The authors warn that a curiosity drive helps only when the change it seeks matches the task at hand.

Catalog record

CREATORJeremias Ferrao, Rafael Fernandes Cunha
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-21047

Browse the full library · About the Tower of Babel