Trade-offs in planner representation for automated software testing

This paper explores an alternative approach to automated test generation based on Al technology: planner based test generation. In planner based test generation, test cases are generated from a domain theory (model of the system being tested), and a description of the initial and desired (goal) state(s) of the system. We identify three key issues and trade-offs in the process of developing a planner based automated test case generator: how to extract knowledge about the domain (domain analysis), how to represent the knowledge (paragmatics of domain representation in the planner) and how to decompose the generation process to exploit the planner's capabilities. We illustrate what we have learned with examples and evaluation of the resulting test cases and the process of generating them.