Finite State Machines (FSM) and Behavior Trees (BT) are widely used for behavior modeling in autonomous intelligent systems. Although they are theoretically equivalent and convertible, existing conversion techniques struggle to preserve behavioral completeness and avoid model complexity explosion. To address these challenges, we present an LLM‑driven unified conversion framework that enables automatic, efficient, and semantically consistent bidirectional transformation between FSM and BT.
Key innovations of the framework are:
-
A novel loop‑execution BT structure that allows the LLM to accurately capture loop constructs in FSM, ensuring full behavioral coverage.
-
A depth‑compression strategy for BT‑to‑FSM conversion, where LLM prompts eliminate redundant control nodes and differentiated hierarchical conversion rules reduce the number of required sub‑FSMs, mitigating state‑explosion issues.
Simulation experiments across various autonomous decision‑making scenarios demonstrate that the proposed framework achieves precise automated conversion and markedly improves scalability and maintainability compared with traditional approaches. It offers a practical solution for behavior model conversion in consumer‑grade autonomous systems such as service robots, game agents, and smart home devices.
Review