World Model Planning for Agents
Methods for training LLM agents to use internal world models for planning and reasoning over long horizons.
19 papers · 3 months
Where this stands
The written synthesis of this thread is for subscribers. Subscribe.
How this thread developed
June 2026
VLMs improve spatial reasoning by actively imagining novel views with a world simulator
Demonstrates active imagination of novel views using a world simulator to improve spatial reasoning.
11 further papers
Sept 2026 · Cornell University
4D foundation models lack robust visual memory for out-of-view objects
Evaluates 4D foundation models' object permanence and motion continuity, testing whether world models maintain robust internal representations of out-of-view objects.
Sept 2026
Benchmark tests whether pretraining on multi-region neural recordings enables transfer across animals and tasks
Benchmarks transfer of pretrained neural representations across animals and tasks, informing world model generalization in planning.
Sept 2026
Touch-aware world models improve contact-rich dexterous robot manipulation
Extends world models with touch-awareness for contact-rich manipulation, improving planning accuracy in dexterous robot tasks.
Sept 2026 · USTC, TongYi Lab
Executable code gives generative worlds persistent rules and evolving state
Provides a new paradigm for world models by representing rules and state as executable code, enabling persistent and editable simulations.
Sept 2026 · Gaoling School of Artificial Intelligence, Renmin University of China
Editing agent reasoning history boosts long-horizon task performance
Proposes AEWM, a world model that edits agent reasoning history to improve long-horizon task performance.
Sept 2026 · HKUST(GZ), CUHK
Visual action rehearsal helps language models control robot manipulation
Introduces World Action Agent, a visual workspace for rehearsal and correction of robot actions, achieving high success on LIBERO-Pro.
Sept 2026 · Fondazione Bruno Kessler, Carnegie Mellon University
Coding agents synthesize robot planners that generalize to unseen instances
Demonstrates that coding agents can generate reusable robot planners that generalize across many unseen task environments.
8 of 19 papers shown