Given the limited time available for training and increased emphasis on self-directed learning in the military, it is essential to develop methods to improve training effectiveness with minimal impact to instructor resources. Training practitioners have attempted to achieve this through the incorporation of automated systems such as Intelligent Tutoring Systems to augment instructor time by emulating human tutors. However, these systems have yet to reach training effectiveness levels that rival those of human tutors. A review of the literature indicates a significant share of the performance gap between computer-based tutoring and human tutors lies in the ability of the humans to be aware of and responsive to the learner's cognitive/affective states. Even so, human tutors have only limited perception of the trainee's cognitive/affect states. When instantiated in a training system, perceptive abilities may allow computer-based tutors to go beyond the abilities of human instructors. It is thus imperative that the trainee model within these systems incorporate both the trainee's performance as well as a diagnosis of their affective and cognitive state.

This paper presents a theoretical framework for the creation of a trainee model that incorporates affective and cognitive state of trainees based on inputs from low-cost, non-intrusive sensors. This framework has theoretical foundations in learning science and physiological measurement and could drastically increase the diagnostic capability of current intelligent training systems. Implementation of this framework could transform adaptive training based on cognitive/affective states from a cost prohibitive endeavor to a goal well within reach. It is hypothesized that a trainee model based on lower cost sensors will account for a significant portion of the variance measured by benchmark sensors/systems that prove expensive or invasive.