Nicholas Byrum · Research Project

AI vs Human Incident Command in VR Fire Training

ResearchVirtual RealityHuman FactorsLLM AgentsFirefighter Training

VR residential basement fire scenario — street view and window approach — The VR residential fire environment used in the study — a low-visibility, high-pressure basement fire scenario.

Goal

Evaluate whether a large-language-model-based AI agent could serve as an incident commander in a virtual reality residential basement fire scenario and compare how firefighter communication changed when interacting with AI versus human command.

Challenges

The main challenge was creating an AI command system that remained operationally realistic, stayed focused on fireground decision-making, and fit naturally into the pace of radio communication without disrupting scenario flow.

Solution & Results

I contributed to the development and evaluation of a VR firefighter training environment that simulated a residential basement fire under low-visibility, high-pressure conditions. The project integrated a large-language-model-based incident commander into a radio-style communication system, allowing direct comparison between AI-led and human-led command.

Radio-style communication transcript between firefighters and the incident commander — Radio-style communication between firefighters and the incident commander.

We analyzed firefighter transcripts from multiple central Iowa departments and compared communication patterns across both conditions. Results showed that AI command prompted more explicit tactical reasoning, clarification-seeking, and projection of next actions, while human command produced shorter status updates and smoother task progression.

Bar chart: prevalence of communication behaviors, AI IC vs Human IC — Prevalence of communication behaviors under AI vs human incident command.

Bar chart: reporting and reasoning patterns, AI IC vs Human IC — Reporting and reasoning patterns across the two conditions.

The project strengthened my experience in research design, human factors analysis, VR-based simulation, and evaluating how AI influences decision-making in safety-critical environments.