During my thesis work I was dealing with robot soccer, where the players constitute a multiagent system. Here to perform successfully we need the effective coordination and cooperation of our autonomous agents which can be analysed in the subject of robot soccer. My specific aim was to build a framework which is suitable to the official environment of the RoboCup competition’s simulation league, where a simulator provides us the model of the physical evironment with the help of a central server. This gives us such an abstraction that allows us to disregard certain areas that are not important in this case, such as object recognition or the physical movements of our robot.
During the semester I’ve been working together with Tibor Oláh. Examining the problem we came to a solution that it would be benefitial to divide the task into two subproblems, the development of a low level strategy (LLS) and a high level strategy (HLS). The LLS realizes a motion planning level and the communication to the server. The HLS is on a higher strategic level, it realizes the team coordination and controls the cooperation. My task in this semester
was to prepare this level.
First of all I examined the convention system of the robot soccer and the way a full game looks like. I divided the behavious of the team into two parts: an administrative one which describes what should we do when the game stops (for example in the case of a free kick or a goal), these were the first steps I realized. The other important part is the so-called HLS, which is built on three different, well-separated column: formation logics, ball handling and the tactics
of the goalie.
The formation logic tells us where should the players position themselves on the field. Similarly to real-life soccer I ordered roles to every agent, where every role has its own areas and tasks.
The part of the ball handling tells us what should a player do if he receives the ball. I defined a graph which tells us what kind of directions can a player with a given role pass the ball forward to. Between the different branches of the graph the choice is non-deterministic considering the different probabilities of the different actions.
The goalie tactic describes the behaviour of the goalie. This mainly concentrates on the optimal positioning and it also handles certain situations when the goalie has to leave its position (the goal area).
To realize the HLS I choose a fuzzy control to work with. First of all I fuzzyfied the input values. I composed the fuzzy variables from the data symbolizing the inner state of the players and sensable part of the field, that I received from the LLS. Later on I tuned the rule system according to the actual game situation. As a summary I can say that I could realize a high level strategy.