Avaliação do uso da Linguagem PDDL no planejamento de missões para robôs aéreos

Abstract

Unmanned Aerial Vehicle (UAV) mission planning is a complex task that is comprised of, among other things, determining which vehicles should be used and which tasks each vehicle has to perform in order to accomplish the desired outcome. In some cases, this task can be too complex for human operators. Using computational tools for mission planning is desirable or even essential in some cases. Automated Planning is the area of AI that develops planning methodologies and techniques to automatically generate the sequences of actions necessary to solve problems in different domains. One important tool in this area is the Planning Domain Definition Language (PDDL). Being domain independent, this language can be applied to problems of distinct nature, from simple blocks worlds to complex logistics problems where time and resources are fundamental dimensions. The present work studies the use of PDDL for UAV mission planning. PDDL is used to automatically generate the sequences of actions necessary to perform two distinct missions developed in this work. The first mission focuses on UAV mobility. We explore some essential aspects such as flight duration, speed, distance and fuel consumption. We then modelthese elements using PDDL to assess how powerful the language is. We develop an experimental framework based on a flight simulator in order to run the PDDLgenerated missions and compare the plans to the simulated reality. This allows us to refine and improve the models and to further explore PDDL and its limitations. The second mission models a forest fire suppression scenario where we explore PDDLs ability to generate temporal plans for multiple UAVs.