Utilize este identificador para referenciar este registo:
https://hdl.handle.net/1822/63766
Título: | Path integral learning of multidimensional movement trajectories |
Autor(es): | André, João Santos, Cristina Costa, Lino |
Palavras-chave: | Path Integral Dynamic Movement Primitives Parametrized Policies Reinforcement Learning Robotics Black Box Optimization |
Data: | 2013 |
Editora: | AIP Publishing |
Revista: | AIP Conference Proceedings |
Citação: | André, J., Santos, C., & Costa, L. (2013, October). Path integral learning of multidimensional movement trajectories. In AIP Conference Proceedings (Vol. 1558, No. 1, pp. 1025-1028). American Institute of Physics. |
Resumo(s): | This paper explores the use of Path Integral Methods, particularly several variants of the recent Path Integral Policy Improvement (PI 2 ) algorithm in multidimensional movement parametrized policy learning. We rely on Dynamic Movement Primitives (DMPs) to codify discrete and rhythmic trajectories, and apply the PI 2 -CMA and PI BB methods in the learning of optimal policy parameters, according to different cost functions that inherently encode movement objectives. Additionally we merge both of these variants and propose the PI BB -CMA algorithm, comparing all of them with the vanilla version of PI 2 . From the obtained results we conclude that PI BB -CMA surpasses all other methods in terms of convergence speed and iterative final cost, which leads to an increased interest in its application to more complex robotic problems. |
Tipo: | Artigo em ata de conferência |
URI: | https://hdl.handle.net/1822/63766 |
ISBN: | 9780735411845 |
DOI: | 10.1063/1.4825679 |
ISSN: | 0094-243X |
Versão da editora: | https://aip.scitation.org/doi/abs/10.1063/1.4825679 |
Arbitragem científica: | yes |
Acesso: | Acesso aberto |
Aparece nas coleções: | DEI - Artigos em atas de congressos internacionais |
Ficheiros deste registo:
Ficheiro | Descrição | Tamanho | Formato | |
---|---|---|---|---|
1.4825679.pdf | 263,49 kB | Adobe PDF | Ver/Abrir |