Generalized Linear Integer Numeric Planning

Anonymous

Keywords: Generalized planning, Numeric planning, Program synthesis, Regular Expressions

Abstract: Classical planning aims to find a sequence of actions that guarantees goal achievement from an initial state. The representative framework of classical planning is propositional logic. Due to the weak expressiveness of propositional logic, many interesting real-world applications cannot be described as a classical planning problem. Some extensions such as numeric planning and generalized planning are therefore proposed. In this paper, we focus on a generalized version of numeric planning, namely generalized linear integer numeric planning (GLINP), requiring each numeric variable to be an integer, and initial states to be formalized as a numeric formula that represents possibly infinitely many states. GLINP is a more expressive planning formalization than qualitative numeric planning. In addition, we develop an approach to synthesize solutions to GLINP problems. This approach generates a solution which can satisfy all instances of the domain as long as the set of initial states is representative. Finally, we evaluate our approach on several benchmarks, and experimental results demonstrate the effectiveness and scalability of our proposed approach.