Agent-Based Assistance in Ambient Assisted Living Through Reinforcement Learning and Semantic Technologies

Abstract

For impaired people, the conduction of certain daily life activities is problematic due to motoric and cognitive handicaps. For that reason, assistive agents in ambient assisted environments provide services that aim at supporting elderly and impaired people. However, these agents act in complex stochastic and indeterministic environments where the concrete effects of a performed action are usually unknown at design time. Furthermore, they have to perform varying tasks according to the user’s context and needs, wherefore an agent has to be flexible and able to recognize required capabilities in a certain situation in order to provide adequate, unobtrusive assistance. Hence, an expressive representation framework is required that relates user-specific impairments to required agent capabilities. This work presents an approach which (a) describes and links user impairments and capabilities using the formal, model-theoretic semantics expressed in OWL2 DL ontologies, (b) computes optimal policies through Reinforcement Learning and propagates these in an agent network. The presented approach improves the collaborative, personalized and adequate assistance of assistive agents and tailors the agent-based services to the user’s missing capabilities.