Skip to Main content Skip to Navigation
Theses

Apprentissage de représentations sensori-motrices pour la reconnaissance d'objet en robotique

Abstract : For several years now, mobile robotic is trying to escape from the amniotic space of research laboratories to explore the unpredictable and even hostile world of our places of living and working, to serve and entertain us. However, conventional logical and symbolic methods of Artificial Intelligence require models of the robot, its actions and perceptions, which must be designed a priori. They are therefore ill-suited towards the unexpected and newness. On the other hand, artificial learning systems, often inspired by biological ones, seem to be on track to provide these first approaches missing ability to adapt. In this thesis, we are considering learning as a central mechanism of the robotic architecture. This architecture can be represented in the face of a sensori-motor loop where actions and perceptions come together in an associative structure. In this architecture, learning allows the acquisition of new knowledge on the environment but is also involved in the modeling of the robot's actions: in a direct manner, by involving combinations of simple instructions on engines, and in an indirect manner, by remembering the effects of these actions on the environment or on the robot itself. In the studied architecture, this form of learning is supported by a pulsed neuron network with unsupervised learning. The architecture we are studying can also express the motivations and goals of the robot through a second learning mecanism, by associating reward values with actions or perceptions by means of reinforcement learning. Thus, the usefulness of each action, their expected reward, are finally exploited by the decision-making process.
Document type :
Theses
Complete list of metadatas

Cited literature [45 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00283073
Contributor : Arlette Evrard <>
Submitted on : Thursday, May 29, 2008 - 9:30:36 AM
Last modification on : Friday, January 10, 2020 - 9:08:08 PM
Long-term archiving on: : Friday, May 28, 2010 - 8:34:53 PM

Identifiers

  • HAL Id : tel-00283073, version 1

Citation

Nicolas Do Huu. Apprentissage de représentations sensori-motrices pour la reconnaissance d'objet en robotique. Automatique / Robotique. Université Paul Sabatier - Toulouse III, 2007. Français. ⟨tel-00283073⟩

Share

Metrics

Record views

290

Files downloads

2048