A Human-Aware Decision Making System for Human-Robot Teams

Abstract

In a human-robot team, robots can perceive the surrounding environment states using various sensors, but typically do not perceive the humans’ internal states (e.g., workload, fatigue, and comfort). This work presents a human-aware system of systems that incorporates human states into the robot’s decision-making process to achieve more fluid human-robot team dynamics and improve the overall team performance. The human-aware system architecture employed a reinforcement learning paradigm as the decision-making agent that incorporated task and human state information. The architecture was validated on teaming data from the NASA MATB-II task environment. The results suggest that the learned action strategies were fine-tuned to the human.