Thinking with Patterns: Breaking the Perceptual Bottleneck in Visual Planning via Pattern Induction

arXiv 2026 51.7 method

TLDR

A training-free visual planning method using Thinking with Images to build internal world models, improved by Pattern Induction for efficiency.

Reasoning

The paper introduces a novel approach to visual planning by leveraging TWI to construct internal world models, with Pattern Induction reducing computational overhead. Strengths include a training-free strategy and empirical validation across multiple domains; weaknesses include potential scalability issues and lack of comparison to RL-based methods.

Read-first score

Read-first score 51.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 29.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

80.9

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.80931874

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=evaluation,experiment,result

Topical relevance 29%

41.4

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 257.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

interactive world model

world simulator

generative world model

video world model

Deep Analysis

Innovations

Formulating Thinking with Images (TWI) as a tool to gradually build and reflect an accurate internal world model for visual planning.
Pattern Inference: a novel TWI strategy that enables VLMs to actively recognize known visual patterns in new tasks and directly infer local world model structures.
Pattern Induction: an online inductive learning strategy that treats visual patterns as composite and reusable experts, autonomously discovered and optimized from experience.

Methodology

The approach uses TWI to iteratively acquire local visual evidence and build an internal world model for planning. To improve efficiency, Pattern Inference recognizes known visual patterns to directly infer world model structures, and Pattern Induction discovers these patterns via online inductive learning from experience. The method is training-free for planning and evaluated on FrozenLake, Crafter, and CubeBench domains.

Key Results

Experimental evaluations in FrozenLake, Crafter, and CubeBench domains show that the proposed approaches achieve a desirable balance between accuracy and efficiency.

Limitations

Too many TWI operations significantly increase computational overhead.

Resources

Paper