Skip to Main content Skip to Navigation

Regroupement de compétences robotiques en compétences plus générales

Abstract : The discovery of sensorimotor contingencies and their structurationinto skills are both important topics in the field of robotics. In particular,robots need the ability to learn skills which are both semantically rich and as general as possible. During this thesis, we studied the question of merging robotic skills into more general skills. After formally defining the notions that can be found in the litterature of skills and parameterized skills, we established a link between paramaterized skills and inverse models, then mirrored the dualism between forward and inverse models to propose a dual type of parameterized skills:forward-parameterized skills. We went on to determine when merging skills into a forward-parameterized skill is relevant and when it’s not. The problem is then formulated as a regression problem, and algorithms inspired by the Occam Razor principle are proposed to find a mixture of experts that solves it with minimal complexity. Those algorithms are then applied to simulated object-manipulation data.
Document type :
Complete list of metadata

Cited literature [62 references]  Display  Hide  Download
Contributor : ABES STAR :  Contact
Submitted on : Monday, October 15, 2018 - 2:59:06 PM
Last modification on : Wednesday, May 11, 2022 - 12:06:07 PM
Long-term archiving on: : Wednesday, January 16, 2019 - 3:19:40 PM


Version validated by the jury (STAR)


  • HAL Id : tel-01895789, version 1


Adrien Matricon. Regroupement de compétences robotiques en compétences plus générales. Robotique [cs.RO]. Université de Bordeaux, 2018. Français. ⟨NNT : 2018BORD0081⟩. ⟨tel-01895789⟩



Record views


Files downloads