Automatic hierarchical nesting of partially observable markov decision processes for task planning in service robotics
A wide variety of approaches have been proposed to address the problem of task planning in robotics, from which partially observable Markov decision processes (POMDP) stand out due to their capacity to model the uncertainty of actions and keep track of the state of the world by means of a partially...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2019 |
| País: | México |
| Institución: | Instituto Nacional de Astrofísica, Óptica y Electrónica |
| Repositorio: | Repositorio Institucional del INAOE |
| Idioma: | inglés |
| OAI Identifier: | oai:inaoe.repositorioinstitucional.mx:1009/1949 |
| Acceso en línea: | http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/1949 |
| Access Level: | acceso abierto |
| Palabra clave: | info:eu-repo/classification/Inspec/Task planning info:eu-repo/classification/Inspec/Hierarchical POMDPs info:eu-repo/classification/Inspec/Service robotics info:eu-repo/classification/Inspec/Declarative programming info:eu-repo/classification/Inspec/General architecture info:eu-repo/classification/cti/1 info:eu-repo/classification/cti/12 info:eu-repo/classification/cti/1203 info:eu-repo/classification/cti/120323 |
| Sumario: | A wide variety of approaches have been proposed to address the problem of task planning in robotics, from which partially observable Markov decision processes (POMDP) stand out due to their capacity to model the uncertainty of actions and keep track of the state of the world by means of a partially observable representation of it. Nonetheless, there are some drawbacks inherent to the use of POMDPs, such as designing a representation that models as best as possible a particular problem, along with the complexity that represents to find a good policy for POMDPs with large state spaces. Therefore, in order to mitigate these challenges, in this thesis we propose an architecture for task planning oriented towards service robot applications, that combines a knowledge representation scheme and POMDPs to build a hierarchy of actions that enables the decomposition of problems into several smaller ones. The knowledge representation defines a list of parameters, so that domain specific information can be encoded by a designer, and used by the architecture to automatically generate and execute plans to solve tasks. Using the hierarchy of actions to generate plans, the system is able to exploit the structure of the environment and ignore those regions in the state space that are irrelevant for a specific task. To evaluate the proposed architecture, a mobile robot navigation domain is employed as case study. Experimental results show that, in scenarios with moderate uncertainty, the architecture is able to perform both reliably and time efficiently, as it generates plans in a time that is several orders smaller than baseline methods. |
|---|