Strategic Implementation of Super-Agents in Heterogeneous Multi-Agent Training for Advanced Military Simulation Adaptability
IEEE Access, cilt.13, ss.96544-96563, 2025 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 13
- Basım Tarihi: 2025
- Doi Numarası: 10.1109/access.2025.3573419
- Dergi Adı: IEEE Access
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, Directory of Open Access Journals
- Sayfa Sayıları: ss.96544-96563
- Anahtar Kelimeler: Training, Artificial intelligence, Scalability, Adaptation models, Decision making, Complexity theory, Multi-agent systems, Military computing, Heuristic algorithms, Computer architecture, Advanced combat simulations, AI-driven tactical decision making, heterogeneous multi-agent systems, military simulation training, reinforcement learning, proximal policy optimization, super-agents and tactical scenario analysis
- Boğaziçi Üniversitesi Adresli: Evet
Özet
This study focuses on the application of reinforcement learning in tactical military simulation environments involving heterogeneous multi-agent systems. Optimizing Heterogeneous Multi-Agent Training (HMAT) through scenario-specific adjustments to the Proximal Policy Optimization (PPO) algorithm, we tackle the complexity of tactical simulations. Utilizing an advanced simulation platform, a diverse range of Reinforcement Learning (RL) agents are rigorously trained across various combat scenarios. A ‘super-agent’, an Artificial Intelligence (AI) orchestrator for multi-agent systems, marks a significant advancement in collaborative AI, enhancing operational performance. Comparative analysis highlights the strengths of both traditional RL approaches and HMAT in a unified computational framework. While independent learning agents excel in predictable environments with fast training capabilities, HMAT stands out in dynamic scenarios for its adaptability and superior performance. The integration of HMAT with ‘super-agents’ is shown to markedly improve the fidelity and adaptive capacity of military simulations. Experimental results demonstrate that our fine-tuned super-agent framework achieves up to 92% mission success rate, outperforming scenario-specific baselines by 15–20% in complex Suppression of Enemy Air Defenses (SEAD) and air-to-ground tasks.These enhancements have far-reaching implications, potentially revolutionizing strategic military training and operational planning and underscoring AI’s critical role in modern defense strategies.