Abstraction through clustering: complexity reduction in automated planning domains

Automated planning is a very active area of research within artificial intelligence. Broadly this discipline deals with the methods by which an agent can independently determine the sequence of actions required to successfully achieve a set of objectives. In this paper, we will present work outlining a new approach to planning based on clustering techniques, in order to group states of the world together and use the fundamental structure of the world to lift out more abstract representations. We will show that this approach can limit the combinatorial explosion of a typical planning problem in a way that is much more intuitive and reusable than has previously been possible, and outline ways that this approach can be developed further.