Orchestrated Reality: From Role-Play to Living, Playable Game Worlds -- LLM-Driven World Simulation as a Parameterized-Action POMDP
Researchers have developed a framework called "Orchestrated Reality" to enable large language models (LLMs) to create persistent, simulated game worlds. This system treats the game world as a canonical object managed by a central orchestration agent, analogous to a Game Master (GM) in tabletop role-playing games. The framework formalizes LLM-driven game worlds as a Parameterized-Action POMDP, where state is represented by JSON entities and transitions are managed by an LLM-driven Plan-Diff-Validate-Apply (PDVA) pipeline. AI
IMPACT Enables LLMs to create persistent, simulated game worlds by formalizing game state and transitions.