A Neuro-Symbolic Approach to Strategy Synthesis for Strategic Logics

2026-06-16

AI Summary

This work addresses the practical intractability of strategy synthesis in multi-agent systems under strategic logics such as Alternating-time Temporal Logic (ATL), which suffers from high computational complexity. The paper proposes the first neurosymbolic framework that integrates a large language model (Qwen3-32B) as a strategy-generation oracle with a formal model checker, enabling efficient search and certification of valid strategies through a generate-and-verify loop. By incorporating the large language model into the strategy synthesis pipeline for strategic logic—while preserving formal correctness—the approach substantially improves solving efficiency. Evaluated on a newly constructed benchmark dataset, NatATL, comprising 4,211 instances, the method achieves 92% accuracy on bounded strategic reasoning tasks.

Abstract

Reasoning about what agents can achieve through strategic interaction is a core challenge in Multi-Agent Systems (MAS). Logics for strategic ability, such as ATL, provide rigorous methods, but their adoption is often hindered by the computational cost of strategy synthesis. We introduce a neuro-symbolic framework that integrates large language models (LLMs) into the model-checking pipeline for MAS. The LLM acts as a strategy-generation oracle, proposing candidate strategies that are then formally validated by a standard MAS model checker. This generate-and-certify architecture uses LLM guidance to navigate large combinatorial strategy spaces while preserving formal soundness: generated strategies are accepted only when certified by the verifier. We instantiate the framework for bounded strategic reasoning in NatATL and introduce the first NatATL strategy-synthesis dataset, consisting of 4211 instances. Experiments with an open-weight Qwen3-32B model show that our certified pipeline achieves 92\% accuracy on strategy-synthesis outcomes.

Problem

• strategy synthesis
• strategic logics
• multi-agent systems
• computational complexity
• formal verification

Innovation

• neuro-symbolic
• strategy synthesis
• large language models
• model checking
• strategic logics