FLEXIBLE SENSING SYSTEMS

Nature, 2019, doi: 10.1038/s41586-019-1234-z

Humans can sense, weigh and grasp different objects, deduce their physical properties at the same time, and exert appropriate forces – a challenging task for modern robots. Studying the mechanics of human grasping objects will play a supplementary role in visual-based robot object processing. These tools require large-scale tactile data sets with high spatial resolution. However, there is no large human-grasped tactile data set covering the whole hand, because dense coverage of the human hand with tactile sensors is challenging. Hence, the capability of observing and learning from successful daily human-object interactions is the long-term goal of aiding the development of robots and prosthetics.

A scalable tactile glove (STG) and deep convolutional neural networks with 548 sensors uniformly distributing over the hand has been fabricated to be used to identify individual objects and explore the typical tactile patterns by Subramanian Sundaram from Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, MA, USA. In their work, the STAG can record tactile videos, measuring normal forces in the range 30 mN to 0.5 N. More importantly, the device can be built with low-cost materials ($10) and used for a long time. They also record a large-scale tactile data set with 135 000 frames, while interacting with 26 different objects. The interaction between these groups and different objects reveals the key correspondence between different regions of human hands when manipulating objects. This as-fabricated artificial simulation lens of the natural mechanoreceptor network provides insight into the tactile characteristics of human beings, thus contributing to the design of prosthesis, robot grasping tools and human-computer interaction in the future.

Zheng Lou (Institute of Semiconductors, CAS, Beijing, China)

doi: 10.1088/1674-4926/40/7/070202