Stretchable Electronic Skin for Tactile Action Recognitions Assisted by Machine Learning

Monitoring tactile pressure and action recognition are important functions of electronic skin (e-skin). To detect pressure distribution by attaching e-skin over three-dimensional objects conformably, e-skins need to be stretchable while maintaining output stability. To address this challenge, we develop a resistive-type tactile pressure sensor using laser-induced graphene (LIG) and liquid metal (GaInSn) patterned on a stretchable ecoflex film. To imitate a simple robotic sphere hand, a balloon structure integrated with a tactile pressure sensor array is demonstrated, allowing that the sensor can also be characterized as a function of stretching conditions by changing air pressure. Finally, echo state network (ESN), which is a reservoir computing framework, is applied for action recognition based on continuous tactile pressure changes and distribution results.<br/>The fabrication process is briefly explained. First, ecoflex solution was poured onto a mold to make liquid metal channel patterns embedded in the ecoflex film and cured in an environmental oven at 90 °C. Laser-induced graphene (LIG) was formed on a PI film via CO₂ laser exposure. Ecoflex solution was then poured over the LIG/PI film, followed by LIG transfer to the ecoflex film after curing it for 15 min in the oven at 90 °C. These two ecoflex films were bonded using ecoflex solution as glue. After lamination, GaInSn liquid metal was injected into the channels to form the electrodes. Finally, the injection ports were sealed with ecoflex solution to complete sensor fabrication.<br/>The real-time pressure sensor output and hysteresis show that the sensor has relatively stable operation during the application and release of pressure. The electrical resistance behaviors by applying pressure up to 140 kPa have two trends for low and high pressures. At pressures less than 80 kPa, the sensor sensitivity was 2.73 %/kPa, while the sensitivity was 10.49 % /kPa in the range of 80 ~ 140 kPa. This sensitivity difference is most likely caused by the structural deformation differences, which is confirmed by photos during applying pressure. For the stability test, cycle test was conducted by applying 4.3 kPa up to 10,000 cycles over 12 hours, and the results showed constant outputs, suggesting that this sensor has highly stable function. Since it was designed to be a stretchable e-skin, it was verified that the sensor could be stretched more than 200 % without electrical disconnection. The stabilities of the sensor against stimuli of partial surface damage, temperature change, humidity change, and pH change were also confirmed. After confirming stable operation as a stretchable tactile pressure sensor, a balloon-type 4×4 tactile pressure sensor array was fabricated to control the degree of expansion of the e-skin by injecting air. As a proof-of-concept, three actions of patting, sliding, and grabbing were performed with/without an air injection volume of 80 mL of the e-skin. By optimizing ESN algorithm, tactile pressure distribution and action recognition were successfully realized with average accuracy of &gt;85 % regardless of stretching conditions.<br/>In summary, a stretchable e-skin with good durability was demonstrated by optimizing and designing the structure and materials. In the future, as a practical application of electronic skin, it is expected to reduce the cost of device fabrication by optimizing the design and applying it to human-computer interaction by developing machine learning algorithms.