Test sequences as plans: an experiment in using an AI planner to generate system tests

Most automated software testing techniques adopt a procedural approach, specifying how sequences are to be generated. In contrast, AI planners generate plans (sequences of actions) in a goal oriented fashion, specifying how actions interact and inferring sequences from that information. We describe the development of an prototype test generation system which is built in an AI planner. We demonstrated the prototype on an application with a command language interface adopted from the Sleuth system, a tool for automated application domain testing.