Risk-Aware Order Dispatching–Grabbing Framework for Cross-Domain Swarm Task Allocation

Existing heterogeneous multi-agent task assignment methods often assume static or homogeneous platforms, limiting their applicability in dynamic air-sea combat scenarios. This paper proposes a priority-aware hybrid scheduling framework that integrates large language model (LLM)-based threat assessment, conditional value-at-risk (CVaR)-driven spatial clustering, and auction-based allocation. High-priority targets are assigned via centralized dispatch to ensure rapid response, whereas medium- and low-priority clusters are managed through decentralized auctions to reduce system load and enhance resource utilization. Evaluations conducted in a three-dimensional (3D) air-sea simulation environment demonstrate efficient coordination between unmanned aerial vehicles (UAVs) and unmanned surface vehicles (USVs), effectively handling up to 140 dynamic targets while reducing assignment latency and improving overall mission efficiency. The results indicate that the proposed framework provides a scalable and practical solution for cross-domain cooperative operations under uncertainty.