Proceedings of the Annual Meeting of the Cognitive Science Society

This paper aims to answer the fundamental but still unan-swered question: how can brains represent 3D objects? Ratherthan building a model of visual processing, we focus on mod-eling the haptic sensorimotor processes through which objectsare explored by touch. This idea is inspired from two mainfacts: 1) in developmental terms, tactile exploration is the pri-mary means by which infants learn to represent object shapes;2) blind people can also represent and distinguish objects justby haptic exploration. Therefore, in this paper, we firstly es-tablish the relationship between the geometric properties of anobject and constrained navigation action sequences for tactileexploration. Then, a neural network model is proposed to rep-resent 3D objects from these experiences, using a mechanismthat is computationally similar to that used by hippocampalplace cells. Simulation results based on a 2 × 2 × 2 cube anda 3 × 2 × 1 cuboid show that the proposed model is effectivefor representing 3D objects via tactile exploration and compar-ative results suggest that the model is more efficient and accu-rate when learning a representation of the 3×2×1 cuboid withan asymmetrical geometrical structure than the 2 × 2 × 2 cubewith a symmetrical geometrical structure.