Hye-rim Han, Jong-geun Choi, Journal of Internet Computing and Services, Vol. 26, No. 4, pp. 89-97, Aug. 2025
10.7472/jksii.2025.26.4.89, Full Text:  HTML
Keywords: MUM-T, Multi-agent, Command & Control System, Decision-Making system, UAV, GCS, sLLM

Abstract

Manned-Unmanned Teaming (MUM-T) is a vital defense technology, enabling collaboration between manned and unmanned platforms across diverse domains. However, current MUM-T systems heavily rely on heuristic approaches, increasing the risk of decision-making errors and inconsistencies in complex battlefield scenarios. This study addresses these challenges by proposing an sLLM-based multi-agent system for MUM-T task planning. The key contributions include: (1) defining UAV reconnaissance data and building a virtual MUM-T dataset, (2) implementing a multi-agent based pipeline and interactive interface for real-time task planning generation and delivery, and (3) designing sLLM-based agents to minimize errors and maintain security. The proposed system demonstrates the potential of sLLM in defense applications, offering improved task planning accuracy and reliability for military operations. By applying the system to 600 virtual scenarios, it achieved high scores in the BLEU and ROUGE evaluations, with an average score of 0.95 or above, thereby demonstrating the accuracy and consistency of its task plan outputs. This confirmed that its performance was comparable to that of ChatGPT 4o.

Statistics
Cite this article