FactorSmith：通过马尔可夫决策过程分解生成自主模拟

Generating executable simulations from natural language specifications poses significant challenges due to the limited reasoning capabilities of large language models (LLMs) when dealing with extensive interconnected codebases. FactorSmith presents a framework that synthesizes playable game simulations in code from textual descriptions by integrating two complementary concepts: factored POMDP decomposition for principled context reduction, and a hierarchical planner-designer-critic agentic workflow for iterative quality refinement at each generation step. Drawing from the factored partially observable Markov decision process (POMDP) representation introduced by FactorSim, the proposed method effectively decomposes simulation specifications to manage their complexity. This approach aims to enhance the quality and playability of the generated simulations.