Proceedings of the Annual Meeting of the Cognitive Science Society

Teamwork in Team Space Fortress, a real-time cooperative task, was studied by analyzing the performance of participantspaired with different partners. To defeat the fortress, a player taking the role of bait approaches within the fortress rangeof fire causing the fortress to lower its shield to fire, thereby becoming vulnerable to attack by a partner playing therole of shooter. A novel design exchanging partners within four person groups allowed the identification of adaptationsand isolation of individual contributions to team performance. Team performance was determined by factors at bothindividual and team levels. Using subjective similarity rankings collected on Amazon Mechanical Turk, we constructedhigh-dimensional embeddings of similarity between team trajectories. Results showed that team members who adaptedmost, led to improved team performance. In re-pairings of partners better individual performance did not necessarily leadto better team performance again supporting the need for adaptivity in human machine teaming.