Neural Network Mapping and Clustering of Elastic Behavior From Tactile and Range Imaging for Virtualized Reality Applications

To fully reach its potential, virtualized reality needs to go beyond the modeling of rigid bodies and introduce accurate representations of deformable objects. This paper explores neural networks and vision-based and tactile measurement strategies to investigate the intricate processes of acquisition and mapping of properties characterizing deformable objects. An original composite neural network framework is applied to guide the tactile probing by clustering measurements representing uniform elasticity regions and, therefore, direct sensors toward areas of elasticity transitions where higher sampling density is required. The network characterizes the relationship between surface deformation and forces that are exemplified in nonrigid bodies. Beyond serving as a planner for the acquisition of measurements, the proposed composite neural architecture allows the encoding of the complex force/deformation relationship without the need for sophisticated mathematical modeling tools. Experimental results prove the validity and the feasibility of the proposed approach.