An LLM-Based Agentic Natural Language Dialogue System Approach for Helicopter MUM-T Operations

In this work, we present our approach for integrating a large language model (LLM)-based assistance agent into a military helicopter simulator to enable natural language interaction during Manned-Unmanned Teaming (MUM-T) scenarios. While extant simple voice command interfaces can reduce pilot workload, the use of LLM agents in military aviation, particularly in mission-relevant applications, remains underexplored. To the best of our knowledge, this is the first attempt to suggest an LLM-based agent in this domain. After introducing the overall challenge, we establish the need for such an agent by sharing insights from an experimental campaign involving experienced military pilots. We describe the operational domain, a hierarchical multi-agent architecture, designed to support flexible interaction while minimizing hallucinations. We qualitatively evaluate the integration of certain preliminary components into our helicopter research cockpit simulator. The system needs to be fully tested in an evaluation study in order to prove the concept.