Intégration de la navigation et de la sélection de l'action dans une architecture de contrôle inspirée des ganglions de la base

Abstract : The functions of action selection and navigation are essential for the design of adaptive robots. Action selection concerns the choice, at all time, of the most appropriate behaviour that maintains survival. This choice depends on the environmental context, and on the internal states and motivations of the robot. Navigation deals with locomotion, mapping, localization and path planning. The integration of both these abilities is crucial for an autonomous robot, for instance to retrieve hidden resources by using path planning. Up to now, few engineering models have focused on interfacing them. Recently, models of the basal ganglia --vertebrate neural structures, implicated in the integration of spatial information for action selection, have been proposed. The aim of our work has been to build from this neurobiological knowledge an architecture for action selection, which is able to integrate sensorimotor, motivational and spatial information. We have first adapted an existing biomimetic model of action selection and tested its ability to solve a survival task in a robotic implementation. The comparison with a simple «winner-takes-all» mechanism has demonstrated that the dynamical properties of the model provides adaptive capacities, e.g. limitation of behavioural dithering, maintenance of internal variables at higher levels and reduction of energy consumption Drawing inspiration from the distinct roles of the dorsal and ventral circuits of the basal ganglia --resp. action selection and navigation integration, we have then elaborated an architecture interfacing this model with two navigation strategies: object approach and topological navigation . A simulated robot, tested in a similar survival task, has been able to use both path planning and object approach to reach distant resources and exploit unknown supply, to cope with various internal states and environmental configurations, and to survive in a large environment where all the previous abilities had to be exhibited. We came to the conclusion that the basal ganglia circuits would provide a robust system interfacing action selection and navigation for autonomous robots. However, some complementary neurobiological knowledge would be necessary to refine the biological plausibility of our model. Adding reinforcement learning --processes that implicate the basal ganglia-- is necessary to enhance the adaptivity of our architecture.
Complete list of metadatas

Cited literature [195 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00007683
Contributor : Benoît Girard <>
Submitted on : Tuesday, December 14, 2004 - 2:50:08 PM
Last modification on : Monday, August 26, 2019 - 2:08:01 PM
Long-term archiving on : Friday, April 2, 2010 - 9:24:50 PM

Identifiers

  • HAL Id : tel-00007683, version 1

Citation

Benoît Girard. Intégration de la navigation et de la sélection de l'action dans une architecture de contrôle inspirée des ganglions de la base. Neurosciences [q-bio.NC]. Université Pierre et Marie Curie - Paris VI, 2003. Français. ⟨tel-00007683⟩

Share

Metrics

Record views

466

Files downloads

785