Paper notes

Web Agents with World Models: Learning and Leveraging Environment Dynamics in Web Navigation

Hyungjoo Chae, Namyoung Kim, Kai Tzu-iunn Ong, Minju Gwak, Gwanwoo Song, Jihoon Kim, Sunghwan Kim, Dongha Lee, Jinyoung Yeo

First arXiv submission: October 17, 2024 · ICLR 2025

Predicts the effects of candidate web actions as natural-language descriptions of important changes, then uses those predictions for action selection.

Read paper ↗

Contribution and setting

Introduces a transition-focused abstraction that reduces the burden of predicting long, repetitive web observations and supports world-model-augmented planning.

Model output

Natural-language descriptions of changes in the next web observation.

Action interface

Candidate web-navigation actions.

State and memory

The planner uses web context and predicted descriptions of state changes.

Relationship to NeuralOS

Connects environment prediction to agent decisions in web navigation.

Source ↗

Scope and limitations

The paper evaluates web navigation using textual descriptions of state changes.

Related terms: web world model, counterfactual planning, transition prediction, WMA.

Read or reuse this entry

JSON record · Markdown · Find in the full list