From Force Control and Sensory-Motor Informations to Mass Discrimination

Abstract

AbstractHuman adults know that usually, big objects are heavier than small ones if these objects are quite similar, in the same material for example. They have a general idea of the weight affordances about the every-day life objects. This paper presents a neural network architecture coupled with a simple linear actuator using force control, designed to use sensory-motor and visual informations during manipulation to learn how to recognize objects of different masses. After learning the association of sensory-motor informations through time with a particular object, our architecture can discriminate different masses and give relevant information for unknown objects, consequently, the objects are associated to some of their inferred physical properties.KeywordsForce ControlNeural Network ArchitectureWeight PerceptionHeavy ObjectSpectrum GeneratorThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.