DeepAgent Ph II

SoarTech proposes DeepAgent, a system for automatically learning and improving multiple, realistic behaviors in diverse, complex environments using observations of human gameplay. On the Phase II effort, SoarTech will transition the DeepAgent system developed on Phase I to a more realistic first-person shooter training environment. The base period will focus on addressing challenges introduced by the more complex simulation. At the end of the base period, we will demonstrate DeepAgent controlling fire team or squad level teams in scenarios similar to those played in military virtual training exercises. During the option period, SoarTech will transition to a third-party environment with many active users and focus on improving behaviors through the use of human examples. Data from these games will be collected and used to continuously improve DeepAgent’s behaviors, which will be used to play against human opponents in a live version of the simulation.