Each human fingertip has more than 3,000 touch receptors that largely respond to pressure. Humans rely heavily on sensation in their fingertips when manipulating an object, so the lack of this sensation presents a unique challenge for individuals with upper limb amputations. While there are several dexterous prosthetics available today, they all lack the sensation of “touch.” The absence of this sensory feedback results in objects inadvertently being dropped or crushed by a prosthetic hand.
To enable a more natural feeling prosthetic hand interface, researchers incorporated stretchable tactile sensors using liquid metal on the fingertips of a prosthetic hand. Encapsulated within silicone-based elastomers, this technology provides key advantages over traditional sensors including high conductivity, compliance, flexibility, and stretchability. This hierarchical multi-finger tactile sensation integration could provide a higher level of intelligence for artificial hands.
The researchers used individual fingertips on the prosthesis to distinguish between different speeds of a sliding motion along different textured surfaces. The four different textures had one variable parameter: the distance between the ridges. To detect the textures and speeds, researchers trained four machine learning algorithms. For each of the ten surfaces, 20 trials were collected to test the ability of the machine learning algorithms to distinguish among the ten different complex surfaces comprised of randomly generated permutations of four different textures.
Results showed that the integration of tactile information from liquid metal sensors on four prosthetic hand fingertips simultaneously distinguished among complex, multi-textured surfaces, demonstrating a new form of hierarchical intelligence. The machine learning algorithms were able to distinguish among all the speeds with each finger with high accuracy.
The team compared four different machine learning algorithms for their successful classification capabilities: K-nearest neighbor (KNN), support vector machine (SVM), random forest (RF), and neural network (NN). The time-frequency features of the liquid metal sensors were extracted to train and test the machine learning algorithms. The NN generally performed the best at the speed and texture detection with a single finger and had a 99.2 percent accuracy to distinguish among ten different multi-textured surfaces using four liquid metal sensors from four fingers simultaneously.
Although advances in prosthetic limbs have been beneficial and allow amputees to better perform their daily duties, they do not provide them with sensory information such as touch. They also don’t enable them to control the prosthetic limb naturally with their minds. The new technology could help provide amputees with a more natural prosthetic device that can “feel” and respond to its environment.
For more information, contact Gisele Galoustian at