Orchestrated Reality: From Role-Play to Living, Playable Game Worlds -- LLM-Driven World Simulation as a Parameterized-Action POMDP

arXiv 2026 59.2 method, application

TLDR

Proposes an LLM-driven game world framework formalized as a Parameterized-Action POMDP with a singleton orchestration agent and JSON state representation.

Reasoning

Strengths include a novel formalization addressing persistent world state in LLM-driven games and a clear architectural separation. Weaknesses are that it is work in progress with no empirical evaluation or real-world benchmarks, limiting evidence of effectiveness.

Read-first score

Read-first score 59.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 6%

100

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

Citation impact 18%

94.7

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.94704179

Methodology quality 18%

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

Topical relevance 29%

64.3

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

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: 459.

Keyword Scores

world model

world simulator

interactive world model

world dynamics prediction

generative world model

model-based reinforcement learning world model

video world model

Deep Analysis

Innovations

Formalization of LLM-driven game world as a Parameterized-Action POMDP with state as a tree of canonical JSON entities
Singleton orchestration agent (Game Master) that owns the world as a canonical object
Plan-Diff-Validate-Apply (PDVA) pipeline for committing schema-validated, content-hashed JSON deltas
Separation of narrative projection from underlying canonical state to address state persistence issues

Methodology

The framework models the game world as a Parameterized-Action POMDP where state is a tree of canonical JSON entities, actions are decomposed into a discrete intent kind and structured JSON parameters, and the agent observes only a narrative projection of state. The transition kernel is implemented via a Plan-Diff-Validate-Apply (PDVA) pipeline that commits schema-validated, content-hashed JSON deltas. The paper provides a formal model, a JSON-state example, a worked single-turn example, and a catalogue of 15 illustrative incidents from a real deployment.

Key Results

No empirical results are reported; the paper presents a work-in-progress framework with illustrative incidents from a real deployment. Empirical validation through a planned human player study, multi-NPC concurrent agency, and deployment as an RL environment are situated as future work.

Limitations

No empirical validation or quantitative results are provided; only illustrative incidents from a real deployment are shown
The framework is described as work in progress, with future work needed for human player studies, multi-NPC concurrent agency, and RL environment deployment
Today's deployed LLM-driven systems do not sustain a persistent world with validated state representation, and the proposed framework is presented as an architectural choice rather than a proven solution
The system relies on LLM-driven pipelines which may introduce unpredictability or errors in state transitions

Resources

Paper