This paper describes work being performed under Phase II of a Small Business Innovative Research Project (SBIR) for the Army Research Institute. A goal of this effort is to improve the realism of computer-generated force CGF entities in constructive simulations. Currently, CGF entity behavior is very predictable and unrealistic with respect to the natural variability with which humans perform given varying amounts of aptitude, training, and environmental stressors to which they would be exposed on a real battlefield.

A second goal of the SBIR effort is to develop a set of algorithms and data structures for including variables such as aptitude, training, and stressor effects that can be integrated with other types of available software packages used for developing human performance models.

The work is currently progressing on three main thrusts. One thrust is focused on developing the aptitude algorithms, learning curves, and stressor algorithms that will eventually influence the performance variables in the human performance models. Work in support of this thrust included a review of the literature on individual and team learning theories, military requirements for human performance modeling and a search of the literature for empirical data that describes the effects of training on human performance. Work toward developing the learning curves has also included the development of a data collection questionnaire for obtaining estimates from soldiers on how their training affected their proficiency in combat performance. This questionnaire was administered to platoon leaders and platoon sergeants from Armor divisions at Fort Riley, Kansas and Fort Carson Colorado. Data from the questionnaires has been analyzed and used to develop learning curves for classroom, simulator, and field training effects.

A second thrust of the project is the development of a software tool that will allow a user to enter information about the training and aptitude of a population of operators in a human performance model. This information will be used as input to the learning curves to calculate the appropriate changes to performance variables in the human performance models. In order to make the tool generalizable for any type of training or human performance model, we are designing it to be completely configurable by the user. This feature of the software tool will allow users that have data to generate their own learning curves use those learning curve algorithms to affect the performance variables in the models. A part of this effort was the development of a test bed model that we are using to test and demonstrate the functionality of the software tool in correctly modifying performance variables.

The third major thrust of the project is to develop the architecture for communicating performance variable values between the software tool and the human performance models. Included in this effort is the selection of an appropriate entity-based constructive simulation such as ModSAF, OneSAF testbed, or JointSAF to apply the tool described above. At this time, the eventual target platform is the OneSAF test bed. However, since all three of these simulations share the same code base, we have begun this work on the JointSAF software. One of the major challenges in developing the communication architecture is in identifying the performance and potential stressor variables in the simulation that can be modified. This paper presents details of the work that has been and work that is yet to be done for each of these three major thrusts.