患者の見守り支援システムにおける深層強化学習を用いた転倒転落リスク低減方策決定方法についての研究

抄録

The fall of the patients in the hospitals are one of the main causes of prolonged hospital stay. However, it is difficult for all clinical staff to judge and act at the expert level immediately in aa situation where the condition of patients and environment changes at any time. Therefore, as a clinical safety system that assists clinical staff for an aging society with a low birthrate, we propose and develop a nursing assistant system using artificial intelligence and robot technology . In this paper, we present a study n the design automation of deep reinforcement learning model mounted on the autonomous mobile robot assisting to reduce patients' fall risk. First, This method observes state transitions, and actions of veteran clinical staff(experts). Nest, estimates reward functions and action vectors using deep inverse reinforcement learning. finally, deep reinforcement learning performs using these, and the reward function and the action vectors are updated. The results show that not only reward definitions that were designed by humans, but also action definitions are automated. It also suggests that intervention strategies, such as the use of "leverage principle" behind veteran clinical staff behavior, can be acquired. This study is useful for expert behavior and also for human resource education.