State Aggregation through Reasoning in Answer Set Programming

For service robots gathering increasing amounts of information, the ability to realize which bits are rel- evant and which are not for each task is going to be crucial. Abstraction is, indeed, a fundamental characteristic of human intelligence, while it is still a challenge for AI. Abstraction through machine learning can inevitably only work in hindsight: the agent can infer whether some information was per- tinent from experience. However, service robots are required to be functional and effective quickly, and their users often cannot let the robot explore the environment long enough. We propose a method to perform state aggregation through reasoning in an- swer set programming, which allows the robot to determine if a piece of information is irrelevant for the task at hand before taking the first action. We demonstrate our method on a simulated mobile ser- vice robot, carrying out tasks in an office environ- ment.