Large Language Models for Verifiable Sequential Decision-Making in Autonomous Systems
Keywords: Large Language Model, Sequential Decision-Making, Formal Method, Verification
TL;DR: We develop an algorithm that constructs automaton-based representation encoding the high-level task knowledge from large language models and provide a method for formally verifying the task knowledge.
Abstract: Automaton-based representations of task knowledge play an important role in control and planning for sequential decision-making problems. However, obtaining the high-level task knowledge required to build such automata is often difficult. Meanwhile, large language models (LLMs) can automatically generate relevant task knowledge. However, the textual outputs from LLMs cannot be formally verified or used for sequential decision-making. We develop a novel algorithm named GLM2FSA, which constructs a finite state automaton (FSA) encoding high-level task knowledge from a brief natural-language description of the task goal. The proposed algorithm thus fills the gap between natural-language task descriptions and automaton-based representations, and the constructed FSA can be formally verified against user-defined task specifications. We accordingly propose a method to iteratively refine the queries to the LLM based on the outcomes, e.g., counter-examples, from verification. We demonstrate GLM2FSA's ability to build and verify automaton-based representations of everyday tasks and also of tasks that require highly specialized knowledge.
Submission Number: 4
Loading