Using Inductive Rule Learning Techniques to Learn Planning Domains.

When dealing with complex problems, designing a planning domain becomes a hard task that requires time and a skilled expert. This issue can be a problem when trying to model a planning domain intended to work in real-world applications. In order to overcome this problem, domain learning techniques are developed aiming to learn planning domains from existing real-world processes. Domain learning techniques then must face typical problems from this kind of applications such as data incompleteness. In this paper, we extend a classification algorithm developed by our research group, in order to create a highly resistant to incompleteness domain learner. We achieve this by extracting the information contained in a collection of plans and creating datasets, applying cleaning and preprocessing techniques to these datasets and then extracting the hypothesis that model the domain's actions using the classifier. Seeking a first validation of our solution before trying to work with real-world data we test it using a collection of simulated standard planning domains from the International Planning Competition. The results obtained shows that our approach can successfully learn planning actions even with a high degree of incompleteness.