Finally, team coordination is difficult to achieve in the face of the possibility that autonomous team members may not agree on the <time-stamped-team-strategy> or the mapping from teammates to roles within the team strategy. Again, there are no disastrous results should team members temporarily adopt different strategies; however they are more likely to achieve their goal G if they can stay coordinated.
One method of coordination is via the locker-room agreement. Agents agree on globally accessible environmental cues as triggers for switches in team strategy. Another method of coordination which complements this first approach is via the time stamp. When hearing a message from a teammate indicating that the team strategy is different from the agent's current idea of the team strategy, the agent adopts the more recent team strategy: if the received message's team strategy has a time-stamp that is more recent than that on the agent's current team strategy, it switches; otherwise it keeps the same team strategy and informs its teammate of the change. Thus changes in team strategy can quickly propagate through the team.
In particular, suppose that agent heard at time t that the team formation is . It then hears a message from agent indicating that the team formation was set to at time u. If t<u, then is a more recent team decision and it updates its notion of the team's formation to . However, if u<t, it is agent that has an obsolete view of the formation. then sends a message to indicating in the <time-stamped-team-strategy> field that the formation was set to at time t, thus causing to update its notion of the team's formation. In the rare even that t = u, the locker-room agreement must specify an order of precedence among roles in order for the agents to determine which agent's idea of the current formation to regard as correct.
Depending on the available bandwidth in the particular application, the <selected-internal-state> can also be used to facilitate team coordination by helping to keep the team members up-to-date regarding the mapping , and perhaps regarding object locations that might be hidden to individual agents.