Decision Making

As the environment of the soccer game is very complex, every situation encountered by the players should be evaluated and the best possible action taken. An example of an action would be moving to a point or passing the ball. To evaluate a game state a cost function is programmed into every robot. The cost function is a sum of products of a robot-situation-criterion (rsc) and a weight assigned to given criterion. The higher the weight of the criterion the more "expensive" the cost is. After the cost functions for every player are evaluated, the managing mechanism selects a player take action. Examples of basic rsc:

• rsc1 -distance from assigned field (defense players have to stay in the defined field near the goal). For defense players high deviation would get a high weight while for forward players an advance on the opponent's goal gets a low weight.

• rsc2 - distance to ball. Two teammates can be located close to the ball so the lower weight will be assigned to the closest robot. See Figure .

• rsc3 -freedom of motion. The decision is made when moving from one point to another based on whether there are objects to be avoided. A high congestion of opponents would get a high weight. See Figure .

Figure: (left) Distance s2 is less than distance s1, so JayBot2 is assigned a lower weight for rsc2; (right) here exists a situation where rsc2 for JayBot1 has lower weight than JayBot2. While path1 is shorter than path2, objects need to be avoided, so rcs3 weight forJayBot1 increases.

At the beginning of the training period the cost functions for all robots are evaluated for different situations and the robots rely on the managing mechanism. As the course of training progresses the cost functions for situations encountered are memorized and the reliance on the manager eases.