ActionSplice: In-Flight Action Editing for Interactive World Models

arXiv 2026 34.4 method

TLDR

ActionSplice enables in-flight action editing in chunk-autoregressive video world models via counterfactual state transport, avoiding rollback and improving speed and image metrics on benchmarks.

Reasoning

The paper presents a novel inference-time framework for interactive world models with strong empirical gains and maintains frozen models, which is a practical strength. However, the abstract lacks details about method complexity and limitations, and the evaluation is restricted to specific benchmarks and image-quality metrics.

Read-first score

Read-first score 34.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 44.

Recency 5%

100

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

Topical relevance 25%

62.9

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=evaluation

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%

10.9

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0031

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: 295.

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

ActionSplice inference framework that formulates in-flight action editing as Counterfactual State Transport (CST).
A lightweight corrector transports the interrupted backbone-native representation toward the matched state induced by the revised action at the same solver step, keeping the world model and sampler frozen and resuming without replaying completed evaluations.
Two CST variants: CST_R updates the entire active chunk, while CST_T preserves a temporal prefix and updates only the suffix.

Methodology

The method is an inference framework for chunk-autoregressive video world models. It inserts a lightweight corrector that maps an interrupted backbone-native representation to a counterfactual state matching a revised action at the current solver step, while the world model and sampler remain frozen. Evaluation uses minWM-Wan Action2V and HY-WM1.5 with LPIPS, PSNR, SSIM, rollback-relative comparisons, and pixel-ready speedups versus waiting.

Key Results

CST_R reduces rollback-relative LPIPS by 61.5% on minWM-Wan Action2V and 75.9% on HY-WM1.5 relative to direct condition swapping; CST_T reduces suffix LPIPS by 56.1% and 77.5%, respectively, with 2.73x and 1.69x pixel-ready speedups over waiting. Under the HY-WorldPlay protocol, CST_R achieves PSNR 25.66 dB, SSIM 0.6902, and LPIPS 0.1337 against the original rollout.

Resources

Paper