A novel robot-task-description for a variety of dynamic behaviors

Akihiko Yamaguchi, Tsukasa Ogasawara

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

This paper aims to unify the 'robot language' approach and the reinforcement learning (RL) framework in order to design behaviors of robots with a simple description. We develop a kind of robot language where we describe a robot task, then the robot employs an RL method to acquire the corresponding behavior. The remarkable feature of this approach is that we do not have to specify the procedure of the behavior, and the models of the environment and the robot. To accomplish this approach, we employ the C++ RL library SkyAI as the base system, then we extend the SkyAI's script interface so that we can describe tasks simply. In this mechanism, a task is described with several event-driven functions where the reward and the end-of-episode condition are defined. As the demonstration, we design six kinds of behaviors for a humanoid robot; a crawling, a handstanding, a jumping, a forward rolling, a backward rolling, and a turning task.

Original languageEnglish
Title of host publication2012 IEEE/SICE International Symposium on System Integration, SII 2012
Pages86-92
Number of pages7
DOIs
Publication statusPublished - 2012 Dec 1
Externally publishedYes
Event2012 IEEE/SICE International Symposium on System Integration, SII 2012 - Fukuoka, Japan
Duration: 2012 Dec 162012 Dec 18

Publication series

Name2012 IEEE/SICE International Symposium on System Integration, SII 2012

Other

Other2012 IEEE/SICE International Symposium on System Integration, SII 2012
CountryJapan
CityFukuoka
Period12/12/1612/12/18

ASJC Scopus subject areas

  • Information Systems

Fingerprint Dive into the research topics of 'A novel robot-task-description for a variety of dynamic behaviors'. Together they form a unique fingerprint.

Cite this