An Adaptive Genetic-Based Incremental Architecture for the On-Line Coordination of Embedded Agents

This paper proposes a novel embedded agent architecture that aims to coordinate a system of interacting embedded agents in real-world intelligent environments using a unique on-line multi-objective and multi-constraint genetic algorithm. The embedded agents can be complex ones such as mobile robots that would operate hierarchical fuzzy logic controllers or simple ones such as desk lamps that would bear threshold functions instead. The architecture would enable the agents to learn the users' desires and act in real time without the users having to repeatedly configure the system. The multi-embedded-agent system can adapt on-line to handle sudden changes such as unreliable sensors and actuators as well as agents that break down or come into the system. Multifarious experiments were performed on implementations of the aforementioned architecture where the system was tested in different scenarios of varying circumstances, and most importantly on mobile robots and embedded agents in the iDorm-a smart room at the University of Essex.