The paper deals with field-based coordination of agent team in which the continental divide game is applied as a coordination mechanism. The agent team consists of self-interested mobile intelligent agents whose behaviour is modelled using coordination policies based on adaptive learning algorithms. Three types of learning algorithms have been used: three parameter Roth-Erev algorithm, stateless Q-learning algorithm, and experience-weighted attraction algorithm. The coordination policies are analyzed by replicator dynamics from evolutionary game theory. A case study describing performance evaluation of coordination policies according to the analysis is considered.