Abstract
Recent developments in language models have created new opportunities in airtraffic control studies. The current focus is primarily on text andlanguage-based use cases. However, these language models may offer a higherpotential impact in the air traffic control domain, thanks to their ability tointeract with air traffic environments in an embodied agent form. They alsoprovide a language-like reasoning capability to explain their decisions, whichhas been a significant roadblock for the implementation of automatic airtraffic control. This paper investigates the application of a language model-based agent withfunction-calling and learning capabilities to resolve air traffic conflictswithout human intervention. The main components of this research arefoundational large language models, tools that allow the agent to interact withthe simulator, and a new concept, the experience library. An innovative part ofthis research, the experience library, is a vector database that storessynthesized knowledge that agents have learned from interactions with thesimulations and language models. To evaluate the performance of our language model-based agent, bothopen-source and closed-source models were tested. The results of our studyreveal significant differences in performance across various configurations ofthe language model-based agents. The best-performing configuration was able tosolve almost all 120 but one imminent conflict scenarios, including up to fouraircraft at the same time. Most importantly, the agents are able to providehuman-level text explanations on traffic situations and conflict resolutionstrategies.