|
In spite of numerous efforts undertaken to develop processes and procedures for the test and evaluation of the performance and accuracy of computer-generated actors (CGAs), much remains to be done before we can confidently field CGAs that reliably provide a desired suite of behaviors. While it currently appears to be impossible to completely validate CGA behaviors, we believe that an experiment-based approach to CGA behavior evaluation can provide a high degree of confidence in the accuracy of the CGA behaviors and that, as a result, CGA behaviors can confidently be considered to be acceptable within the bounds of the testing and assessments related to its intended use. Clearly, exhaustive testing of all possible CGA behaviors is not feasible since exhaustive testing of CGA behaviors would be an even more computationally complex task than software testing; exhaustive software testing has proven to be impossible for any but the most trivial software. Therefore, our approach to addressing the CGA behavior test and evaluation challenge is based upon the use of the Unified Modeling Language (UML) and the eXtensible Markup Language (XML) to capture and describe the desired CGA behaviors. UML provides the capability to document the desired behavior from a number of perspectives and XML allows us to augment the UML documentation in a standard, open manner. The paper is organized as follows. Section One will contain an introduction to motivate our research and a discussion of the challenges that must be addressed to properly model human behavior and then test and evaluate CGA human behavior models. Section Two will contain a discussion of the relevant background technologies for our work. Section Three contains the discussion of our approach to CGA behavior testing, assessment, and evaluation and how we believe that UML and XML should be used for CGA behavior testing documentation. Section Four contains a summary and suggestions for further research.
|
