Using Artificial Intelligence Planning Methods to Generate Accurate Action Models and Plans to Test Software

Software testing is imperative because even a single error in the code can not only cost a company a lot of money but also its reputation.  For this reason, testing must be exhaustive.  This task can be time consuming and costly if done manually.  The goal of our research is implement artificial intelligence planning methods to generate test cases that emulate user interaction to find most, if not all, errors in the software.  Using PDDL, an AI planning language, we have written domains, which represent the user's available states and actions, and problems, which give the user's initial conditions and goals, so we can then generate plans, a sequence of actions that test the software.  To check the accuracy of our method we will need to compare it to problems, as previously defined, that are generated by inputting domains and plans; we have not yet determined which program we will utilize. To check how thorough the generated plans are, we will input the plans into a program that will give us these results; this program is currently being modified to perform more efficiently.  We intend to see that it is possible to use AI planning methods to accurately depict user sessions on software.