How is a new team strategy during a robocup competition sent to each player of a robot team? Robots in the Standard Platform League (i.e. SPL), for example, are fully autonomous and there is no connection with non-team members (except pulling from the GameControl).
In this leagues, robots can utilize WiFi communication to either broadcast packages to their team members, or listen to GameController messages.
The GameControler provides information about the game state (e.g. ready (go to position), play). Thus, the robots can use the information provided by the GameController to have a strategy according to the game state, or according to the scored goals.
Broadcasted messages by other players reveal how many players may be active / taking part. Moreover, robots can inform other team mates when they get penelized (a robot conducted an illegal action, e.g., pushing another robot).
There are more dynamical strategies used in the SPL. According to the fact whether one robot is manipulating the ball, other robots may aim for a specific position in the vicinity (supporting).
Much work is done, but since most strategies are dependent on the WiFi coverage/robustness, just a few dynamic strategies are used so far during games, sicne WiFi reliability turns out to be low during competitions. That means, packages may need several seconds to arrive. Even the chances to play a game without WiFi are high (happens partially to the SPL finals in 2014/2015). Most teams therefore tend to use static behaviour (e.g. players have a static position and react when the ball is in their vicinity).
Therefore, some teams aim to have a strategy WiFi independent (partially relying on vision) to overcome this gaps.
Short answer here: Most times, strategy is derived from communicated messages over the WiFi.
For further reading, I would recommend the SPL team description papers under the link provided.