Show Me Your Moves

As electronics continue to rapidly shrink in size and in terms of their energy budget, wearable applications for devices are becoming more and more desirable. The computational power and storage that used to sit on our desktops can now come along with us on our wrists, on an armband, or even in our eyeglasses. These are some good targets for inserting computing devices, but there is another that is perhaps even more natural and transparent to us in our daily lives that has been largely neglected to date — our clothing.

This area will not remain neglected much longer if work like what has recently come out of a lab at MIT continues. A team there has been experimenting with creating smart textiles that tightly conform to the wearer’s body. Using novel techniques, they have been able to produce seemingly normal articles of clothing that are capable of sensing the wearer’s posture and motions. And unlike previous attempts to accomplish this goal, the sensors woven into the fabric operate with a high degree of precision.

To accomplish their feat, the team used a digital knitting process that weaves together standard yarns with functional yarns. A conductive yarn is knit into a grid, and is sandwiched between piezoresistive materials. Where these materials intersect, a pressure sensor is created. Changes in pressure can be detected by observing variations in the resistance level of the piezoresistive material that occur when the fabric is squeezed. While this approach works, there is a great deal of noise – which reduces sensor accuracy – injected into the signal as layers of the material rub against one another.

The team’s major insight that alleviated this problem was to incorporate a special type of plastic yarn into the knitting process. When lightly heated, this plastic yarn will melt and fuse the layers of the fabric together, while also giving the garment a more snug, form-fitting style. It is this characteristic that greatly reduces sensor noise and allows for highly accurate measurements to be captured.

There was still work to be done to turn those pressure measurements into any kind of useful information, however. In order to translate this data into information about the wearer’s posture and motions, the team took a deep learning approach. After scanning the grid of sensors for pressure readings, the data is fed into a convolutional neural network that has been trained to classify various common actions a person might be engaged in.

To test the system, the team created a smart shoe that was designed with their fabrication technique. A yoga mat was similarly constructed. As study participants wore the shoes and performed a variety of yoga poses on the mat, that data was streamed to a computer running the neural network model for real-time classifications. The addition of plastic yarns to the textiles showed their merit here, with an overall classification accuracy rate of 99% being observed.

The researchers see many possible future applications for their technology. They envision shoes being trained to track the gait of someone learning to walk again after an injury, or socks that monitor the pressure on the feet of patients with diabetes to prevent ulcer formation. But they note that by using textiles, there is a lot of flexibility in the design process. That means this process could be adapted to a great many use cases in theory. And since these devices can be seamlessly and comfortably integrated into our lives, that is a future we can all look forward to.