Mobile Robot Planning using Action Language BC with Hierarchical Domain Abstractions

ABSTRACT. Action language BC provides an elegant way of formalizing robotic domains which need to be expressed using default logic as well as indirect and recursive action effects. However, generating plans efficiently for large domains using BC can be challenging, even when state-of-the-art answer set solvers are used. Since a number of task planning domains in mobile robotics can be easily broken up hierarchically using macro-actions, we investigate the computational gains achieved by planning over a hierarchical abstraction of the domain. Each layer of abstraction is described independently using BC, and we plan independently over each description, resolving macro-actions as we go down the hierarchy. We present a case study where at least an order of magnitude speedup was achieved in a robot mail collection task by using hierarchical domain abstractions.