Introducing Actions in CommonSens: A Hybrid Agent-Based Approach

Abstract

Systems that provide ambient assisted living (AAL) are currently emerging at a rapid pace, and the amount of functionalities these systems provide is constantly growing. AAL provides assistance at home for sick, disabled or elderly people, by placing sensors in their homes. These sensors are used for monitoring and assistive purposes.

Activities of daily living (ADLs) are all the possible situations that may occur in the home. Situations we wish to monitor are called events. A complex event processing (CEP) system analyzes the data received from the sensors and detects events. In order for, e.g. the monitored person not waking up in the morning, to be detected by health personnel, CEP systems can send an alarm as a response to each event. A challenge is how to supplement such a system with the ability to respond to events in a more flexible way by acting in the environment, e.g. by turning on the lights and checking if the person wakes up. With actions, the CEP system can act upon the environment, using actuators placed in the home.

Events and actions are dual terms, since both describe a set of states in the environment. Events are predefined changes that may be detected, and actions are planned changes that may be executed upon the environment. Actions can be implemented in CEP systems using intelligent agents. An intelligent agent can be regarded as an entity which observes and acts upon an environment through sensors and actuators.

We have considered how the CEP system CommonSens can be extended to include the functionality of intelligent agents. For describing agents we use the planning domain definition language (PDDL). We propose an architecture that reflects the duality between events and actions. Additionally, we design an agent architecture for CommonSens, and propose the overall design. Properties in the system are chosen with two important ideas in mind: (1) The resulting design will represent the environment and current functionalities in CommonSens as accurately as possible. (2) The level of complexity in the intelligent agent is kept at a minimum in order to gain knowledge and experience with the behavior of agents.

We use a hybrid agent architecture which can handle alarms in a reactive manner, as well as having the functionalities needed for goal-directed behavior. The latter gives an opportunity for the system to support both atomic and complex actions. The hybrid agent architecture consists of three layers. The reactive layer is responsible for sending alarms to health personnel when a critical event has occurred. The action selection layer handles atomic actions. The planning layer constructs plans, which constitute complex actions. We also discuss how the life cycle of intelligent agents in CommonSens will look like.