Learning planning action models with numerical information and logic relationships using classification techniques

Abstract

The task of constructing a planning domain is difficult and requires time and vast knowledge about the problem to be solved. This paper describes PlanMiner-O3 a planning domain learner designed to alleviate this problem, based on the use of a classification algorithm, able to learn planning action models from noisy and partially observed logic states. PlanMiner-O3 is able to learn continuous numerical fluents as well as simple logical relations between them. Testing was realized with benchmark domains obtained from the International Planning Competition and the results demonstrate PlanMiner-O3’s capabilities of learning planning domains.