3D Recognition of Obstacles Using A Vibrissa-like Tactile Sensor

Many rodents are able to detect the shape and texture information of objects [1] - [2] by means of tactile hairs in their snout region (mystacial vibrissae) [3] - [6] . Since these hairs are made up of dead tissue, all sensing has to be performed in the support of each vibrissa, the so-called follicle-sinus complex (FSC) [7] . Guided by the natural paragon we set up a mechanical model, consisting of a cylindrical, one-sided clamped bending rod, which is swept along an object quasi-statically. Due to the contact force the rod undergoes (large) bending deflections in three dimensional space which are described using a non-linear bending theory. The scanning sweep is simulated using a Matlab algorithm in two steps: Firstly, the sweep is simulated to generate the support reactions, which might also be measured using an experimental setup. Secondly, the generated support reactions are used in order to determine a sequence of contact points reconstructing and approximating the object contour. The theoretically generated observables allow to reconstruct contact points on the object within numerical boundaries (error 5·10 -11 ). The object reconstruction reveals that it is useful to scan objects in various orientations, in order to avoid reconstruction gaps.

I. Summary and Motivation

Many rodents are able to detect the shape and texture information of objects [1]-[2] by means of tactile hairs in their snout region (mystacial vibrissae) [3]-[6]. Since these hairs are made up of dead tissue, all sensing has to be performed in the support of each vibrissa, the so-called follicle-sinus complex (FSC) [7]. Guided by the natural paragon we set up a mechanical model, consisting of a cylindrical, one-sided clamped bending rod, which is swept along an object quasi-statically. Due to the contact force the rod undergoes (large) bending deflections in three dimensional space which are described using a non-linear bending theory. The scanning sweep is simulated using a Matlab algorithm in two steps: Firstly, the sweep is simulated to generate the support reactions, which might also be measured using an experimental setup. Secondly, the generated support reactions are used in order to determine a sequence of contact points reconstructing and approximating the object contour. The theoretically generated observables allow to reconstruct contact points on the object within numerical boundaries (error 5⋅10^-11). The object reconstruction reveals that it is useful to scan objects in various orientations, in order to avoid reconstruction gaps.