next up previous
Next: Pre-Planned Set-Plays Up: Teamwork Structure Implementation Previous: Dynamic Switching of Formations

Flexible Positions


As emphasized throughout, homogeneous agents can play different positions. But such a capability raises the challenging issue of when the players should change positions. In addition, with teammates switching positions, a player's internal player-position map tex2html_wrap_inline1554 could become incorrect and/or incomplete. The locker-room agreement provides procedures to the team that allow for coordinated role changing. In CMUnited's case, the locker-room agreement designates an order of precedence for the positions within each unit. Unless their pursuit of the ball leads them from their position, players only switch into a more important position than their current position.

By switching positions within a formation, the overall joint performance of the team can be improved. Position-switching has the potential to save player energy and to allow them to respond more quickly to the ball. However, switching positions can also cause increased player movement if a player has to move across the field to occupy its new position. Players must weight the possible costs and benefits before deciding to switch positions.

Like switching formations, switching positions can change external behaviors via their references to the internal state. In Equation 1, switching positions changes the value of returned by MyPosition(), thus also affecting the value of HomeRange(): the player executing the passive offense external behavior chooses its location from a different range of possible positions.

In addition to having the flexibility to switch to a different position, CMUnited-97 agents also have flexibility within their positions. That is, the external behavior references the internal state to determine a range of possible locations that are determined by the player's current position. However, within this range, the role does not specify the player's precise location. For example, in the passive offense external behavior (equation 1), the player must choose its TargetLocation from within the home range of its current position.

In the CMUnited multi-agent approach, the player positions itself flexibly such that it anticipates that it will be useful to the team, either offensively or defensively. The agents can exercise this flexibility within its external behaviors in two ways:

When marking opponents, agents move next to a given opponent rather than staying at the default position home. The opponent to mark can be chosen by the player (e.g., the closest opponent), or by the unit captain which can ensure that all opponents are marked, following a preset algorithm as part of the locker-room agreement.

When using ball-dependent positioning, the agent adjusts its location within its range based on the instantaneous position of the ball. For example, when the ball is on the same side of the field as the agent, the agent tries to move to a point on the line defined by its own goal and the ball. When the ball is on the other side of the field, the player adjusts its position back towards its own goal.

next up previous
Next: Pre-Planned Set-Plays Up: Teamwork Structure Implementation Previous: Dynamic Switching of Formations

Peter Stone
Thu Dec 17 15:26:44 EST 1998