The basic aim of our study is to develop a metric of team performance that goes beyond assessing the mean performance of the team or its members. We think it is crucial  to take into account the interactions of the team members, when measuring team performance. The necessity for such a metric becomes immediately clear if we consider the example of a team consisting of one excellent and several poorly performing members. Based on the average of the performance of the members or even in some tasks on the basis of team scores, the team could be rated well, although no actual team interaction between the members has taken place.

The  current literature on team performance metrics  includes over 100 models for teams and team interactions. Most of these models are limited to very specific situations and specific types of teams. The methods used to evaluate the teams are usually based upon observational analysis, questionnaires, expert opinion, event based measurements and  team error analysis. There exists few models that offer us metrics that are of a general nature and could be applied across  different situations and types of teams. Our attempt is to develop generalizable, quantitative and objective metrics which can be used by the trainer to evaluate the team's performance.

The experimental data we use is MOUT data (Military Operations in Urban Terrain), collected by our collaborators at the Clemson university. We have data from three teams. Depending on their level of expertise the teams are classified as novices, intermediates and experts.  Each team performs the same room clearing task over a number of trials. The data consists of each team member's spatial coordinates, orientation of head and  weapon,  and heart beat data.

We hypothesize that the team interactions can be represented as the flow along a low dimensional uncontrolled manifold. Variations along this manifold will not affect the perfromance. But if the team trajectories move away from the manifold then the performance of the team suffers. This will then provide us with an objective measure of the team's performance. Furthermore we can also look at learning of the novice and intermediate teams by how the trajectories of these teams converges to the manifold over time.

The attached presentations and movies gives a more detailed picture about our methods and preliminary results.