Arian: A General Architecture for Advisable Agents

Integration of advice is intended to acquire knowledge from an agent. with more comprehensive information. This paper describes Arian, a general architecture for advice integration. Advice is injected into the system using the advice model which is a case in a CBR system. The architecture comprises an advice adapter to model incomming advice as cases of a CBR system. The Arian architecture is a 2-layered structure in which a refiner component in the top layer is used to modify agent behaviors. The paper also discusses how trust is modelled to distinguish between multiple sources of advice including humans. Arian, the champion team of RoboCup'02 Rescue Simulation Competitions, as an example of an implemented agent based on this proposed architecture is given to show that the architecture is well suited for advice integration. Test results are provided to show how Arian advice-taking agents change their behavior based on incomming advice.