Bootstrapping perception with action: a case for sensory prediction - Archive ouverte HAL
Pré-Publication, Document De Travail Année : 2020

Bootstrapping perception with action: a case for sensory prediction

Résumé

Recent developments have argued that perception can emerge from sensation when (and to an extent only when) actions are considered. This problem of Developmental Robotics is studied by drawing from the Sensorimotor Contingencies (SMC) theory. Specifically, an extended formalism for describing sensorimotor interaction is proposed, allowing to account for motor and sensory variations. Said formalism is then used to study the properties of sensory prediction in a naive agent, with two main results. First, spatial contingencies make such a prediction internally computable via a particular class of actions. Second, sensory prediction is shown to capture the structure of motor capabilities of the agent. Both those claims are further illustrated in a simulation. This provides a generalized approach to bootstrap the development of perception for naive agents, which can tackle a larger class of problems.
Fichier principal
Vignette du fichier
root.pdf (13.21 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-02596671 , version 1 (15-05-2020)
hal-02596671 , version 2 (31-10-2020)

Identifiants

  • HAL Id : hal-02596671 , version 1

Citer

Jean Godon, Sylvain Argentieri, Bruno Gas. Bootstrapping perception with action: a case for sensory prediction. 2020. ⟨hal-02596671v1⟩
127 Consultations
123 Téléchargements

Partager

More