LLM-Guided Hybrid Architecture for Autonomous Fire Response: Dialog-Driven Planning in Space and Disaster Missions
Swarnamouli Majumdar, Sonny E. Kirkley, Biswadip Basu Mallik · 2025
Large Language Models (LLMs) such as GPT, BERT, and Zenext AI’s proprietary platform are rapidly evolving as key enablers of semantic reasoning, dialog-based autonomy, and tradespace exploration in safety-critical domains. This paper investigates the integration of LLM agents with autonomous fire-response vehicles for deployment in space habitats and Earth-based disaster environments. Building on Apaza and Selva’s dialog-agent tradespace framework, we present a use case involving a Mars simulation rover equipped with a voice-enabled LLM agent, ORION-FR, designed for fire diagnostics, procedural guidance, and autonomous actuation. We propose a hybrid optimization architecture that combines reinforcement learning and linear programming (RL+LP) to support constraint-aware planning under high-risk conditions. By bridging natural language interaction, mathematical optimization, and edge-based inference, this work advances the use of generative AI for mission resilience in both extraterrestrial exploration and terrestrial emergency response.