Programmable World Model

arXiv 2026 39.4 method, system

TLDR

Programmable World Model decouples persistent state from video generation using executable programs and 3D OBBs, enabling controllable game worlds and outperforming existing interactive video models.

Reasoning

The paper introduces a clear framework for persistent, programmatically controllable world state in video generation, with a new benchmark and strong quantitative results. However, the abstract provides limited details on generalization, scalability, and comparison protocols, and the benchmark may be narrow.

Read-first score

Read-first score 39.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 52.

Recency 5%

100

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

Topical relevance 25%

74.3

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=benchmark,evaluation,result

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: 498.

Keyword Scores

world model

interactive world model

video world model

world simulator

generative world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Programmable World Model framework that decouples world-state evolution from visual observation generation
Natural-language instructions translated into executable programs specifying entity states and state-transition rules
Lightweight engine that maintains an explicit, persistent global world state, including off-screen entities and non-visual attributes
State-augmented 3D oriented bounding boxes as an intermediate representation, deterministically compiled into pixel-aligned spatiotemporal conditioning signals for a pretrained video model
CombatStateBench benchmark for evaluating programmable world models

Methodology

The framework decouples state evolution from visual generation: an agent translates natural-language instructions into executable programs, and a lightweight engine executes those programs to maintain an explicit, persistent global world state. State-augmented 3D oriented bounding boxes and target camera trajectories are deterministically compiled into pixel-aligned spatiotemporal conditioning signals for a pretrained video model used as a generative renderer. Evaluation uses the introduced CombatStateBench and reports Count Accuracy and State Accuracy against existing interactive video world models.

Key Results

On CombatStateBench, the method achieves 94% Count Accuracy and 98% State Accuracy, substantially outperforming existing interactive video world models while supporting coherent long-horizon generation.

Resources

Paper