Publication: Aprendizaje por refuerzo seguro para enseñar a un robot humanoide a caminar más rápido
Loading...
Identifiers
Publication date
2013-07-15
Defense date
2013-07-23
Authors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Enseñar a un robot humanoide a caminar es un problema abierto y
desafiante. Los comportamientos clásicos de caminar habitualmente requieren
la puesta a punto de muchos parámetros de control (longitud de paso,
velocidad, frecuencia, etc). Encontrar una configuración inicial o básica de
estos parámetros no es complicado, pero optimizarla para un objetivo (por
ejemplo, caminar rápido) no es tan sencillo, ya que puede hacer caer al robot
humanoide provocando daños, en caso de una optimización incorrecta.
En este proyecto, se propone usar técnicas de aprendizaje por refuerzo seguro
para mejorar el comportamiento de caminar de un robot humanoide
que permite caminar m as rápido que la configuración predefinida. El aprendizaje
por refuerzo seguro asume la existencia de una política segura que
permite aprender una nueva, la cual se representa con un enfoque basado
en casos. Los algoritmos de aprendizaje por refuerzo seguro aplicados
son PI-SRL (Policy Improvement throught Safe Reinforcement Learning) y
PR-SRL (Policy Reuse for Safe Reinforcement Learning). ________
Teaching a humanoid robot to walk is an open and challenging problem. Classical walking behaviors usually require the tuning of many control parameters (step size, speed, frequency, etcetera). To find an initial or basic confi guration of such parameters could not be so hard, but optimizing them for some goal (for instance, to walk faster) is not easy because, when de ned uncorrectly, may produce the fall of the humanoid, and the consequent damages. In this paper we propose the use of Safe Reinforcement Learning for improving the walking behavior of a humanoid that permits the robot to walk faster than with a pre-de ned con figuration. Safe Reinforcement Learning assumes the existence of a safe policy that permits the humanoid to walk, and probabilistically reuse such policy to learn a new one, which is represented following a case based approach. The Safe Reinforcement Learning algorithms used are, PI-SRL (Policy Improvement throught Safe Reinforcement Learning) y PR-SRL (Policy Reuse for Safe Reinforcement Learning).
Teaching a humanoid robot to walk is an open and challenging problem. Classical walking behaviors usually require the tuning of many control parameters (step size, speed, frequency, etcetera). To find an initial or basic confi guration of such parameters could not be so hard, but optimizing them for some goal (for instance, to walk faster) is not easy because, when de ned uncorrectly, may produce the fall of the humanoid, and the consequent damages. In this paper we propose the use of Safe Reinforcement Learning for improving the walking behavior of a humanoid that permits the robot to walk faster than with a pre-de ned con figuration. Safe Reinforcement Learning assumes the existence of a safe policy that permits the humanoid to walk, and probabilistically reuse such policy to learn a new one, which is represented following a case based approach. The Safe Reinforcement Learning algorithms used are, PI-SRL (Policy Improvement throught Safe Reinforcement Learning) y PR-SRL (Policy Reuse for Safe Reinforcement Learning).
Description
Keywords
Robótica, Robots humanoides, Aprendizaje, Inteligencia artificial, Algoritmos