RAR: Rate-Adaptive Rollout for Time-Consistent Interactive World Models

arXiv 2026 33.9 method

TLDR

RAR enables interactive video world models to roll out at controllable temporal rates via Stride-RoPE and multi-stride sampling, achieving multi-rate consistency, rate switching, and longer horizons.

Reasoning

The paper introduces a clear framework to decouple temporal rate from physical horizon, with Stride-RoPE enabling multi-rate rollout and online rate switching. Strengths include novel methodology and transfer to pretrained models; weaknesses are the lack of explicit real-world dataset details and quantitative results in the abstract.

Read-first score

Read-first score 33.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 49.

Recency 5%

100

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

Topical relevance 25%

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

Methodology quality 15%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=none

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

Frontier

Rank sensitivity

Stability: volatile; rank range: 387.

Keyword Scores

world model

interactive world model

video world model

generative world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

Identifies and formalizes temporal-rate coupling in fixed temporal rate world models, including runtime-rate adaptation and physical-horizon coverage trade-offs
Introduces Rate-Adaptive Rollout (RAR), a framework enabling interactive world models to roll out at controllable temporal rates
Proposes Stride-RoPE to align latent frames sampled at different rates on a shared physical-time coordinate, enabling consistent multi-rate rollout, unseen-rate generalization, and online rate switching without discarding history
Uses multi-stride video sampling to jointly learn fine-grained dynamics and long-horizon evolution

Methodology

RAR combines Stride-RoPE, which aligns latent frames sampled at different rates on a shared physical-time coordinate, with multi-stride video sampling to jointly learn fine-grained and long-horizon dynamics. It is evaluated across action- and camera-controlled environments under matched training budgets and is applied to pretrained world models.

Key Results

Across action- and camera-controlled environments, RAR achieves temporally consistent multi-rate rollout and online rate switching while preserving high-rate generation quality and extending physical horizons under matched training budgets. RAR also transfers to pretrained world models while preserving perceptual quality.

Resources

Paper