Collaboration in teams is essential in robot collectives. In order to achieve goals, individual robots would otherwise not be able to accomplish. In a such a distributed and highly dynamic system, a global coordination might not be possible. In this paper, we analyse static team affiliations, defined at deployment time, and compare its efficiency against dynamic team affiliations generated during runtime using random selection. Since operators might not be able to determine all dynamic aspects of the given environment at the time of deployment, we further propose a novel, goal-aware approach to affiliate each robot with a team. This approach brings together insights from biology, sociology, and psychology. In this novel approach, robots only operate on aggregated information from the network which is potentially changing during runtime. Finally, we also introduce an approach to select a team affiliation during runtime using machine learning techniques. Using 60,000 randomised scenarios, we analyse the efficiency and further discuss the different benefits and drawbacks of the proposed approaches.