There are three points about the kick model of the server that should
be understood before looking at our kicking style. First, a kick
changes the ball's velocity by vector addition. That is, a kick
accelerates the ball in a given direction, as opposed to setting the
velocity. Second, an agent can kick the ball when it is in the
``kickable area'' which is a circle
centered on the player (see Figure 2). Third, the ball
and the player can collide. The server models a collision when the
ball and player are overlapping at the end of a cycle. If there is a
collision, the two bodies are separated and their velocities
multiplied by 
.
As a first level of abstraction when dealing with the ball, all reasoning is done as a desired trajectory for the ball for the next cycle. Before a kick is actually sent to the server, the difference between the ball's current velocity and the ball's desired velocity is used to determine the kick to actually perform. If the exact trajectory can not be obtained, the ball is kicked such that the direction is correct, even if the speed is not.
In order to effectively control the ball, a player must be able to kick the ball in any direction. In order to do so, the player must be able to move the ball from one side of its body to the other without the ball colliding with the player. This behavior is called the turnball behavior. It was developed based on code released by the PaSo'97 team[7]. The desired trajectory of a turnball kick is calculated by getting the ray from the ball's current position that is tangent to a circle around the player (see Figure 3). Note that there are two possible such rays which correspond to the two directions that the ball can be turned around the player. Care is taken to ensure that the ball stays well within the kickable area from kick to kick so that the player keeps control of the ball.
Figure 3: The turnball skill.
Figure 2: Basic kicking with velocity prediction.
The next important skill is the ability to kick the ball in a given direction, either for passing or shooting. The first step is to figure out the target speed of the ball. If the agent is shooting, the target speed is the maximum ball speed, but for a pass, it might be better to kick the ball slower so that the receiving agent can intercept the ball more easily. In this case, the agent must take into account the ball's deceleration over time when determining how hard to kick the ball.
In order to get the ball to the desired speed, several kicks in succession are usually required. By putting the ball to the side of the player (relative to the desired direction of the kick) the agent can kick the ball several times in succession. If a higher ball speed is desired, the agent can use the turnball kicks to back the ball up so that enough kicks can be performed to accelerate the ball.
This skill is very predictive in that it looks at future velocities of the ball given slightly different possible kicks. In some cases, doing a weaker kick one cycle may keep the ball in the kickable area so that another kick can be executed the following cycle. In Figure 2, the agent must choose between two possible kicks. Kicking the ball to position A will result in the ball not being kickable next cycle; if the ball is already moving quickly enough, this action may be correct. However, a kick to position B followed by a kick during the next cycle may result in a higher overall speed. Short term velocity prediction is the key to these decisions.