Decision Making for Affective Agents in Assistive Environments

In this paper, we discuss the Decision Making and Learning ability of Affective Agents to make human-like decisions. This work is in the context of Assistive Living Environments (ALE) applications, where an agent is capable of assisting a human in physical and cognitive rehabilitation through multimodal and adaptive interaction. The goal of this research is to investigate what role multimodality plays in producing a natural and effective interaction using Reinforcement Learning. We propose a hierarchical decision making framework for affective agents doing complex tasks. This framework incorporates an internal reward mechanism to make the learning more efficient.