Regulation Function for Agent Adaptation Issues in Ambient Environment
Abstract
In this work we deal with action selection issues for software agent operating in ambient environment which is highly
dynamic. Unpredictable events may occur and inappropriate actions may damage the software and its environment.
Adaptation ability is a key requirement for such issues. We use multi-agent paradigm to address them. We propose a
regulation function within agent architecture. It is a filter which acts on the stream of behaviour before it becomes
or not action. The aim is to cope with the environmental changes without the need to predict them precisely
at the design-time. To this end, we introduce the Influence-Reaction Model into the agent behaviour management
mechanism. To facilitate its application, we implement the resulting architecture as a Java library called MECA.We
experiment it with an agricultural robot moving through a field.