Generative Artificial Intelligence for Scenario Generation and Communications Analysis

As the capabilities of peer threats improve, the need to expand on the capabilities of Live, Virtual, Constructive (LVC) environments continues to increase. These frameworks provide a controlled environment that allows for the adjustment of training parameters. This ensures the training is both safe and adaptable to various scenarios presented to warfighters, thereby enhancing their readiness. However, manual development and hand-tailoring of red threat behavior and scenarios is a time-consuming process resulting in too few training scenarios to cover the real-world diversity in engagement characteristics like red-threat tactics and postures, blue-side tasking, environmental conditions, and warfighter platform capabilities. Additionally, there is currently no mechanism to promptly evaluate communication, a critical component of mission success, after a training event other than manual analysis and evaluation. In Phase I, Toyon proposes developing a Next Generation Threat System (NGTS)-compatible framework capable of generating mission- and trainee-adaptable red threat behaviors and scenarios populated with a robust set of Deep Reinforcement Learning (DRL)-based red-agents capable of adapting in real time to blue-side warfighter behavior. The designed system will be capable of automating debrief generation in natural language to inform subsequent warfighter-training, red-agent selection, and scenario-generation efforts.