English

General Agentic Planning Through Simulative Reasoning with World Models

Artificial Intelligence 2026-05-22 v3 Computation and Language Machine Learning Robotics

Abstract

What does it mean to plan? Current agentic systems, whether scaffolded workflows or end-to-end policies, rely on reactive decision-making: selecting the next action via a fixed procedure with at most undifferentiated adaptive computation (e.g., chain-of-thought) lacking explicit modeling of future outcomes. This limits generalizability, as each new task demands re-engineering rather than transfer of shared reasoning capacity. Humans, by contrast, plan by mentally simulating consequences of candidate actions within an internal world model, a capacity known as simulative reasoning (System II) that supports flexible, goal-directed behavior across diverse contexts. We argue that simulative reasoning through a world model provides a general-purpose planning mechanism for agentic systems, improving upon reactive policies (System I) by grounding decisions in predicted future states rather than pattern-matched responses. To verify this, we introduce SiRA (Simulative Reasoning Architecture), a goal-oriented architecture instantiating simulative reasoning using an LLM-based world model with natural-language belief states, while remaining model-agnostic. We evaluate across three qualitatively distinct task categories: constrained navigation, multi-hop information aggregation, and general instruction following, in a web-browser environment. Across all categories, simulative reasoning achieves up to 124% higher task completion rates than a matched reactive baseline, and increases constrained navigation success from 0% to 32.2% compared to a representative open-web agent. The persistent advantage across distinct task types suggests the benefit stems from generalizable counterfactual evaluation rather than task-specific tuning.

Keywords

Cite

@article{arxiv.2507.23773,
  title  = {General Agentic Planning Through Simulative Reasoning with World Models},
  author = {Mingkai Deng and Jinyu Hou and Zhiting Hu and Eric Xing},
  journal= {arXiv preprint arXiv:2507.23773},
  year   = {2026}
}

Comments

Winner of Berkeley LLM Agents Hackathon (Fundamentals Track); code available at https://github.com/sailing-lab/sira

R2 v1 2026-07-01T04:28:16.749Z