همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Integrated Reactive-Proactive Multi-Agent Coordination: A SACR Communication Protocol with Strategic Planning Framework

نویسندگان :

Mostafa Mehrabi¹ Monireh Abdoos²

1- Shahid Beheshti University 2- Shahid Beheshti University

کلمات کلیدی :

Multi-agent systems،Signal Acknowledge Communicate Resolve (SACR)،cooperative game theory،large language models،planning،communication protocols

چکیده :

Coordinating autonomous agents under strict constraints without extensive training is challenging. Game-theoretic approaches coordinate effectively but lack planning capabilities, while reinforcement learning requires prohibitive training data. We integrate cooperative game theory with LLM-based planning through SACR (Signal, Acknowledge, Communicate, Resolve), a lightweight protocol for priority-based conflict resolution. The framework uses forward prediction to handle saturation constraints and external interference across planning and execution phases. Dynamic coalition formation enables agents to balance formation maintenance, target acquisition, and threat avoidance simultaneously. Evaluating nine agents across formation, interference, and obstacle scenarios shows SACR eliminates all proximity violations versus 4-82 violations in baselines. With LLM planning, Claude Sonnet 4 with extended thinking achieves 89.1-100% plan adherence, outperforming Gemini 2.5 Pro's 80.9-86.2%. The framework provides interpretable, scalable coordination for constrained multi-agent systems. Implementation code is available at https://github.com/mostafamehrabii/multi-agent-sacr-simulation.