Understanding and avoiding interaction‐based instability in pervasive computing environments

This paper aims to address a fundamental problem related to the interaction of rule‐based autonomous agents in pervasive and intelligent environments. Some rules of behaviour can lead a multi‐agent system to display unwanted periodic behaviour, such as networked appliances cycling on and off.

The paper presents a framework called interaction networks (INs) as a tool to describe and analyse this phenomena. In support of this, and as an aid to the visualisation and understanding of the temporal evolution of agent states, a graphical multi‐dimensional model (MDM) is offered. An instability prevention system (INPRES) based in identifying and locking network nodes is described.

Both IN, MDM and INPRES enable system designers to identify and prevent cyclic instability. The effectiveness of the approach is evaluated using both simulated and physical implementations.

The problem of cyclic instability is strongly related to the number of cycles in the IN associated. It is postulated that high coupling and high number of cycles contributes to the system to self‐lock; however, more research is needed in this direction.

The MDM, interaction benchmark, IN theory, INPRES and intelligent locking offer a practical solution to the problem of cyclic behaviour.

Before this work there was no framework for analysing and eliminating the problem of cyclic instability in rule‐based multi‐agent systems.