Multi-agent dynamic system modelling

Modeliranje višeagentnoga dinamičkoga sustava i razvoj arhitekture agenta s mentalnim stavovima koji posjeduju značajke autonomno orijentiranih jedinki u središtu je istraživanja provedenog u okviru izrade doktorskog rada. Model višeagentnoga dinamičkoga sustava mora uspješno prikazati elemente i odnose u dinamičkom sustavu koji se mijenjaju u vremenu. Vrijeme je eksplicitno uključeno u model višeagentnoga dinamičkoga sustava i model mentalnih stavova agenta. Arhitektura višeagentnoga dinamičkoga sustava je hibridna i dopušta različite vrsta agenata (agente s mentalnim stavovima, proširene agenate s mentalnim stavovima i autonomno orijentirane jedinke). U radu su predstavljeni prošireni agenti s mentalnim stavovima (arhitektura, model mentalnih stavova, proces integracije s autonomno orijentiranim jedinkama i njihov model ponašanja). Predloženi agenti razlikuju se od klasičnih agenata s mentalnim stavovima u tome što koriste autonomno orijentirane jedinke s ciljem prilagodbe agenata na dinamiku sustava. Glavna snaga agenata s mentalnim stavovima je što su inspirirani ljudskim konceptom znanja i promišljanja, a nedostatak je nemogućnost prilagodbe u neplaniranim situacijama, jer ne posjeduju mehanizam za učenje. U ovom radu taj nedostatak nadomješten je uz pomoć autonomno orijentiranih jedinki. One ne posjeduju eksplicitno definirani cilj na razini jedinke, već ga postižu zajedno pri čemu ostavljaju informacije u okolini i vođeni su jednostavnim pravilima ponašanja. Integrirane su u arhitekturu proširenog agenta, a djeluju kada agent ne može kreirati namjeru, odnosno pomažu prilagoditi se na neočekivane situacije. Realizacija mentalnih stavova agenata, promišljanje i ponašanje autonomno orijentiranih jedinki pokazani su na primjerima u višeagentskom simulacijskom alatu.The main objectives of the doctoral thesis are modelling of the multi-agent dynamic system MADS and development of the extended BDI (Belief Desire Intention) agent architecture. The BDI agent i.e. agent with mental attitudes is extended with autonomous oriented entities. The MADS model includes a time parameter, which is crucial for defining the model dynamic. The MADS architecture is hybrid and consists of different types of agents: BDI agents, extended BDI agents and autonomous oriented entities. In this doctoral thesis extended BDI agents are proposed. The architecture of the extended BDI agent is developed, the agent’s mental attitudes model is defined, and the integration process with autonomous oriented entities and their behaviour model is described. The difference between extended BDI agents and classical BDI agents is that former use autonomy oriented entities to adapt on unpredictable situations. The main advantage of BDI agents is that the human concept of knowledge and deliberation inspire them. The disadvantage is that they don’t have learning capabilities. Consequently, they can’t modify their behaviour in unpredictable situations. The solution is an extension of the BDI agents with autonomy oriented entities. They don’t know a global goal, their behaviour is simple, and they mutually communicate through the environment emerging the collective behaviour of the entities. The entities are integrated in the extended BDI agent’s architecture and perform when the agent can’t create an intention. Mental attitudes of the extended BDI agents, their deliberation process as well as a behaviour of autonomous oriented entities are evaluated through examples in multi-agent programmable simulation platform