A state-action neural network supervising navigation and manipulation behaviors for complex task reproduction

Florent d'Halluin (1) , Antoine de Rengervé (2) , Matthieu Lagarde (2) , Philippe Gaussier (2) , Aude Billard (1) , Pierre Andry (2)

1 EPFL - Ecole Polytechnique Fédérale de Lausanne
2 ETIS - UMR 8051 - Equipes Traitement de l'Information et Systèmes

In this abstract, we combine work from [Lagarde et al., 2010] and [Calinon et al., 2009] for learning and reproduction of, respectively, navigation tasks on a mobile robot and gestures with a robot arm. Both approaches build a sensory motor map under human guidance to learn the desired behavior. With several actions possible at the same time, the selection of action becomes a real issue. Several solutions exist to this problem : hierarchical architecture, parallel modules including subsumption architectures or even a mix of both [Bryson, 2000]. In navigation, a temporal sequence learner or a state-action association learner [Lagarde et al., 2010] enables to learn a sequence of directions in order to follow a trajectory. These solutions can be extended to action sequence learning. In this paper we propose a simple architecture based on perception-action that is able to produce complex behaviors from the incremental learning of simple tasks. Then we discuss advantages and limitations of this architecture, that raises many questions.

Mots clés: Gaussian Mixture Models; Learning from demonstration; Neural networks

Domaines: Apprentissage [cs.LG]; Informatique

Florent d'Halluin, Antoine de Rengervé, Matthieu Lagarde, Philippe Gaussier, Aude Billard, et al.. A state-action neural network supervising navigation and manipulation behaviors for complex task reproduction. Proceedings of the tenth international conference on Epigenetic Robotics, Nov 2010, Örenäs Slott, Sweden. pp.165--166. ⟨hal-00551698⟩