next up previous
Next: Experiments Up: Using the Learned Behaviors Previous: Reasoning about action execution

Incorporating the RCF in a behavior

In Section 3.1, the method of using a DT as a part of an RCF is described in detail. However, the RCF is itself not a complete client behavior: it only applies when the ball is within kicking distance. This section situates the RCFs within a complete behavior that can then be used throughout the course of a game.

Table 2: The complete behavior used by the clients in game situations.

The player's first priority is always to find the ball's location (only objects in front of the player are seen). If it doesn't know where the ball is, it turns until the ball is in view. When turning away from the ball, it remembers the ball's location for a short amount of time; however after about three seconds, if it hasn't seen the ball, it assumes that it no longer knows where the ball is [2].

Once the ball has been located, the client can execute its behavior. As described in Section 3.1, each player is assigned a particular position on the field. Unless chasing the ball, the client goes to its position, moving around randomly within a small range of the position. The player represents its position as x,y coordinates on the field.

The client chases the ball whenever it thinks that it is the closest team-member to the ball. Notice that it may not actually be the closest player to the ball if some of its teammates are too far away to see, and if they have not announced their positions recently. However, if a player mistakenly thinks that it is the closest player, it will get part of the way to the ball, notice that another teammate is closer, and then turn back to its position. When the ball is within a certain small range (arbitrarily 10m), the client always goes towards the ball. When the ball is moving towards the client or when a teammate has indicated an intention to pass in its direction, the client watches the ball to see if either of the two above conditions is met. A player that was chasing the ball is predisposed to continue chasing the ball. Only if it finds that it should turn back persistently for several steps does it actually turn back. As required for use of the DT, every player is equipped with the trained Neural Network (see Section 2) which can be used to help intercept the ball.

Finally, every team member uses the same RCF. Whenever the ball is within kicking distance, the client calls its RCF to decide whether to dribble, kick, or pass, and to where. The behavior incorporating the RCFs is laid out in Table 2.

next up previous
Next: Experiments Up: Using the Learned Behaviors Previous: Reasoning about action execution

Peter Stone
Sun Dec 7 06:59:19 EST 1997