Flexible robotic skin has become a hot research topic in recent years because of its soft-touch feature, which resembles living skin. The stretchability of robotic skin could enhance the robot–human experience of touch. In this study, robotic skin was synthesized through additive manufacturing to obtain a soft, flexible, and stretchable sheet achieved via gel extrusion using acidic-type polydimethylsiloxane (PDMS). The integration of carbon nanotubes (CNT) into PDMS can transform flexible silicone from a non-conductive to a conductive material through stable electrical conduction at a CNT mass ratio > 6% w/w. Although carbon nanotubes may degrade silicone flexibility, the residual softness of the composite is still much superior to the rigid force sensor. Testing of the robotic skin involved pressing the robotic finger attached to the Cartesian robot to press onto the robotic skin. Several pressing positions generated various electrical signals from only four wiring contacts at the corners of the robotic skin sample. The electrical signals resulting from CNT deformation within the silicone/CNT composite created a reversible resistance loop while inducing mechanical elongation. Pressing with the robotic finger altered the electrical signals, depending on the different pressing points. The data collected were then used to train convolutional neural networks. High accuracy in position identification was successfully accomplished through training at an F1 score of 0.96, which indicated high precision in the touching prediction. The same neural network technique was also used to evaluate the thermal response on the skin with a lower F1 score of 0.89, indicating that the thermal response of the robotic skin was less accurate than its ability for mechanical identification. This study developed a successful CNT/silicone-based robotic skin model integrating a CNN that demonstrated high sensitivity to touch and heat through 4 contact points, closely mimicking human skin.
Chotiyasilp et al. (Tue,) studied this question.