Specification-Driven Generation and Evaluation of Discrete-Event World Models via the DEVS Formalism
🤖 AI Summary
Existing world models struggle to simultaneously ensure reliability, verifiability, and online adaptability in complex environments governed by discrete events. This work proposes an intermediate approach grounded in natural language specifications: leveraging large language models to iteratively generate discrete-event world models that conform to the DEVS (Discrete Event System Specification) formalism, thereby decoupling model structure from behavior. The generated models are validated against semantic and temporal constraints derived from structured event trajectories, ensuring consistency over long-horizon simulations. This framework supports observable behaviors, reproducible verification, localized diagnostics, and efficient online deployment, offering a principled pathway toward trustworthy and adaptive world modeling in discrete-event settings.
Application Category
📝 Abstract
World models are essential for planning and evaluation in agentic systems, yet existing approaches lie at two extremes: hand-engineered simulators that offer consistency and reproducibility but are costly to adapt, and implicit neural models that are flexible but difficult to constrain, verify, and debug over long horizons. We seek a principled middle ground that combines the reliability of explicit simulators with the flexibility of learned models, allowing world models to be adapted during online execution. By targeting a broad class of environments whose dynamics are governed by the ordering, timing, and causality of discrete events, such as queueing and service operations, embodied task planning, and message-mediated multi-agent coordination, we advocate explicit, executable discrete-event world models synthesized directly from natural-language specifications. Our approach adopts the DEVS formalism and introduces a staged LLM-based generation pipeline that separates structural inference of component interactions from component-level event and timing logic. To evaluate generated models without a unique ground truth, simulators emit structured event traces that are validated against specification-derived temporal and semantic constraints, enabling reproducible verification and localized diagnostics. Together, these contributions produce world models that are consistent over long-horizon rollouts, verifiable from observable behavior, and efficient to synthesize on demand during online execution.
Problem
Research questions and friction points this paper is trying to address.
world models
discrete-event systems
DEVS formalism
specification-driven generation
model verification
Innovation
Methods, ideas, or system contributions that make the work stand out.
DEVS formalism
discrete-event world models
specification-driven generation
LLM-based synthesis
verifiable simulation
🔎 Similar Papers
No similar papers found.
