SCIENTIFIC SYSTEMS CO INC — Department of Defense STTR Phase II: N17A-T025

Description

As systems become more heterogeneous and distributed, determining optimal actions for each “teammate” within the system becomes increasingly complex. A particular difficulty in guiding the actions of these teammates is the fact that they are not guaranteed to follow suggestions provided by a supervisor, especially when those suggestions are inconsistent with the teammate’s tendencies, preferences, or capabilities. We propose the teammate aware autonomy (TAA) system to provide course of action suggestions to each teammate within such a system, where the suggestions (1) are aimed to optimize overall system performance, and (2) are modulated with respect to models of each individual teammate’s tendencies, preferences, and capabilities. The foundational technologies in TAA are a teammate modeling approach to develop models of teammate policies and latent states based on observations, and a multiagent learning algorithm which can optimize multiagent system performance while incorporating these teammate models to ensure suggestions are likely to be understood and accepted by the teammates.