Somebody Pinch Me

When it comes to user interfaces for portable and mobile computing devices, we have to think outside the keyboard. For some use cases, like typing out an email on a smartphone, a virtual keyboard will usually do the trick. But when we are otherwise occupied, perhaps by driving or exploring a virtual world, we cannot give that level of focus to the device. A number of alternative interfaces have been proposed for these scenarios, such as voice and gesture recognition.

Each of these alternatives comes with some annoyances, however. Voice interfaces don’t work well in noisy environments, and they are terrible with regard to privacy. Gesture-based interfaces, on the other hand, just make us look wacky. Nobody really wants to be waving their arms around, “touching” nonexistent objects in front of a crowd. To address these issues, a team led by researchers at the University of Michigan came up with a micro-gesture recognition system called EI-Lite. Wearing a watch-like wristband and making subtle gestures with the thumb and fingers is enough to control a device using this approach.

Unlike traditional gesture recognition systems that rely on cameras or inertial sensors, EI-Lite detects gestures through electrical impedance sensing. This technique measures how the body’s internal tissues respond to small alternating electrical currents. As muscles move and tendons shift, the impedance (resistance to electrical flow) changes in measurable ways. By tracking those tiny variations, EI-Lite can determine what the user’s fingers are doing, even when the movements are imperceptible to the eye.

The system is composed of two main parts: a wrist-worn sensing band and a custom impedance sensing board. The band is lightweight, flexible, and adjustable, designed for comfort during daily wear. It uses four small electrodes — the minimum number needed for what’s called a four-terminal impedance measurement. Each electrode assembly contains a brass contact ball that touches the skin, housed in a 3D-printed PLA case that slides onto a TPU ribbon strap. The arrangement forms a two-by-two grid around the wrist, ensuring that electrical signals can travel through multiple paths in the forearm’s tissues.

Those signals originate from the compact sensing board, which is built around a Teensy 4.0 microcontroller. The board generates a tiny alternating current using a signal generator and a voltage-controlled current source based on a refined Howland circuit. The researchers made several improvements to this circuit to keep it stable and precise, even while rapidly switching between electrode pairs at a 100 Hz sampling rate. A 12-bit digital-to-analog converter controls the signal strength, while high-quality instrumentation amplifiers and buffer op amps capture the resulting voltage changes with minimal noise. The final readings are digitized through an analog-to-digital converter and processed by the microcontroller.

To train the system, the team collected data from 15 participants performing six common micro-gestures, including pinches, releases, and swipes, as well as an idle state. They also recorded continuous pinch force measurements in Newtons, which let their machine learning models estimate how hard the user was pressing. The results were quite encouraging. EI-Lite achieved over 96% accuracy in gesture recognition and predicted pinch forces with a mean squared error of just 0.3 Newtons.

As far as real-world applications are concerned, the researchers demonstrated EI-Lite’s use for AR and VR interfaces, enabling subtle hand control even when the headset’s cameras can’t see the user’s hands. They also showed how it could control presentation slides or assist people with limited hand mobility by turning small, comfortable movements into powerful interaction commands. This almost invisible interface might be just what we need to control the devices of tomorrow.