Representation World Model: Learning States, Transition and Executable Plans in Representation

arXiv 2026 33 method

TLDR

Proposes Representation World Model that learns states, transitions, and executable plans directly in representation space, enabling planning via latent path construction without rollouts or search.

Reasoning

The paper presents a novel approach that integrates planning into representation geometry, avoiding recursive rollouts and action-space search, which is a strong conceptual contribution. However, the abstract provides limited details on experimental settings, baselines, and limitations, making full assessment difficult.

Read-first score

Read-first score 33, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 34.

Recency 5%

100

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

Methodology quality 15%

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

Topical relevance 25%

48.6

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 15%

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

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 252.

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

Introduces the Representation World Model (RWM) that learns states, transitions, and executable plans directly in representation space.
Incorporates planning directly into the learned representation geometry instead of using explicit dynamics models with search, optimization, or policy-based prediction.
Uses inverse-dynamics supervision locally along latent paths constructed from endpoint representations to learn geometry that preserves task-relevant state and transition information.
Enables inference-time planning by directly constructing a latent path between current and goal representations and recovering actions via inverse dynamics, without recursive rollouts or action-space search.

Methodology

RWM learns representation geometry by applying inverse-dynamics supervision locally along latent paths constructed from endpoint representations, requiring these paths to preserve task-relevant state and transition information. At inference, planning directly constructs a latent path between current and goal representations, and inverse dynamics recovers the corresponding actions without recursive rollouts or action-space search. The approach is evaluated on continuous-control benchmarks and robotic manipulation tasks.

Key Results

Experiments on continuous-control benchmarks demonstrate RWM's effectiveness for direct planning, and robotic manipulation results show potential for extension to more complex embodied control tasks. The authors conclude that planning directly in representation space is a promising alternative to conventional world-model planning.

Resources

Paper