Adaptive Task Allocation for Multiple UAVs via LLM-based Strategic Parameter Generation

대규모 언어 모델 기반의 전략적 파라미터 생성을 통한 상황 적응형 군집 무인기 임무 할당

This paper proposes a novel hybrid framework for multiple Unmanned Aerial Vehicle (UAV) task allocation that bridges the semantic gap between strategic intent and numerical control in dynamic environments. Traditional optimization algorithms often fail to adapt flexibly to unpredictable events due to their reliance on rigid, rule-based logic. To address this limitation, we introduce a hierarchical hybrid decision-making architecture that integrates Large Language Model (LLM) with the Consensus-Based Bundle Algorithm (CBBA). In this framework, the LLM serves as a strategic planner, interpreting the semantic context of dynamic events and generating high-level strategic parameters, which then guide the CBBA's decentralized execution. Furthermore, to rigorously analyze the system's performance, we established multi-faceted evaluation metrics encompassing mission success, time efficiency, reallocation, survivability, and resource efficiency. Simulations comparing our method against Greedy and standard CBBA algorithms demonstrate its superior strategic adaptability. The simulation results demonstrate that the proposed method exhibits human-like strategic flexibility. This study validates the LLM as an effective strategic optimizer for resilient swarm operations.