• Cheneler, David
• Pape, L.
• Carrozza, M. C.
• Kazerounian, S.
• Oddo, C. M.
• Kaklamani, Georgia
• Adams, Michael J.
• Anthony, Carl J.
• Bowen, James
• Grover, Liam M.

Abstract

NanoBioTouch is an FP7 funded project that has an overall aim of developing NEMS tactile sensors for integration in an articulated robotic finger. The design of the sensors and signal processing are based on a multidisciplinary approach to improving the current understanding of the human mechano-transduction system. A range of NEMS arrays and bio-NEMS sensor technologies are being designed and fabricated in order to discriminate textures and assess their pleasantness with a resolution that is comparable to that of human subjects. They are being incorporated into a multiphalangeal biorobotic finger with artificial intelligence for enabling discriminative and affective touch. Silicone elastomer is used as the artificial skin with a fingerprint texture and it was found that their spacing relative to the individual sensors was important in generating discriminative textural signals. The current NEMS sensors enable discrimination among surfaces having spatial periods differing down to 40 μm, both under passive-touch and under human-like active-touch tasks. In the case of gratings, this corresponded to an accuracy of > 97.6%. A range of machine learning strategies are being adopted for interpreting the data that includes spatiotemporal phase analysis and a neuromorphic approach to translate the analogue signals into spikes that are similar to those produced by the mechanoreceptors in the human finger pad. In addition, signal processing software has been developed that autonomously learns tactile skills on the robotic finger using a curiosity-driven learning algorithm and that allows real-time motor control and sensor readout. Such curiosity-driven exploration enables the robotic finger to develop tactile skills, by rewarding the finger as when it explores novel methods for recognizing and learning about tactile sensations that it has not previously learnt. Interestingly, this leads to the sequential development of tactics, from the use of tapping motions to more complex sliding motions.Significant progress has also been achieved for the bio-NEMS sensors, which involves the development of the equivalent of the subcutaneous tissue in the human finger pad by using alginate gels. Acellular gels exhibited a strong capacitance change with amplitude that depended on the imposed strain. When a population of live fibroblast cells was encapsulated in such gels there was an additional spiked response with a characteristic time that was believed to be associated with the transport of ions across the cell membranes. This behaviour has some analogies with the action potentials emitted by the mechanoreceptors.

Topics

• impedance spectroscopy
• surface
• phase
• texture
• machine learning
• elastomer