Pub-AI: AI in Science

Driving · 2026-09-30

What Must a World Model Distinguish for Planning?

Rongzhe Wei, Hans Hao-Hsun Hsu, Peizhi Niu, Yifan Li, Pan Li

arXiv:2609.33030PDF

Auto-summarized: this summary was generated by a language model from the paper’s text and has not been reviewed by an editor. Check the paper before relying on it. Benchmark numbers appear only after a human has verified them.

TL;DR

The authors formalize when a world model must preserve physical distinctions for planning, via mechanism/response/decision sufficiency and query-dependent resolution requirements. Experiments on collision, learned nonlinear dynamics, and robotic planning show that needed information depends on the query, candidate set, and planning stage. They also compare query placement: a modular design (query-aware proposal + query-independent action-conditioned prediction) reuses physical predictions across objectives and achieves lower held-out regret than action-conditioned prediction methods.

Why it matters

Planning-quality depends on which distinctions a planner needs: not all predictive detail required for accurate simulation is needed for final selection, especially under coarse decision queries. The work links representation fidelity requirements to planning queries, candidate sets, and intermediate search operations, and evaluates where to inject query information in world-model planning for better generalization across objective compositions.

Method

  • Define mechanism, response, and decision sufficiency as equivalence relations over physical worlds induced by a planning query and candidate set; extend to approximate decisions using response/decision covers and a response–decision gap.
  • Study controlled environments (collision, double-pendulum) to measure how required representation information changes with decision precision, horizon, and planning resolution.
  • Evaluate learned planning systems where query information enters either at candidate proposal only (modular), only in outcome prediction (action-conditioned world model), or coupled in both (world-action model), and test generalization to unseen objective compositions.

Limitation

The authors state that empirical rates depend on the observation process, model class, optimization, coding scheme, and readout, so they characterize the studied systems rather than universal information-theoretic limits; continuous-action analysis isolates an idealized resolution effect approximated by finite candidate sets; robotic studies use simulated Push-T and PushCube environments and adaptive-search results are demonstrated primarily with CEM.

Abstract (from arXiv)

World models simulate the consequences of action candidates, but good planning need not preserve every physical distinction required for accurate prediction. We formalize this gap through a hierarchy of mechanism, response, and decision sufficiency. Given a candidate set, the planning query determines which physical variations matter and how precisely they must be preserved: coarse decisions can discard much of the information needed for prediction, whereas fine decisions may require nearly the same resolution. In practice, planners often adaptively search to construct candidates, and information unnecessary for final selection may still be needed to discover good candidates. What a world model must preserve therefore depends on the query, the candidate set, and the planner. We study these effects in a collision system, nonlinear dynamics, and robotic planning. These varying requirements raise a design question: where should query information enter the planning system? A model that jointly generates actions and outcomes conditioned on the query achieves lower regret than an action-conditioned world model on seen objectives, but this advantage largely disappears when generalizing to unseen objectives. Motivated by this, we propose a modular design in which the query determines where to look and an action-conditioned model predicts what will happen, allowing the same predictions to be reused across objectives.

Related papers