MIT researchers have built a wristband that uses ultrasound imaging and AI to translate wrist muscle movement into precise finger-position data, enabling real-time wireless control of a robotic hand. The device, developed by mechanical engineering professor Xuanhe Zhao and colleagues at MIT and USC, pairs a miniaturized ultrasound transducer with a hydrogel skin adhesive. An AI model, trained on human-labeled ultrasound images, continuously maps the internal state of 34 muscles, 27 joints, and over 100 tendons into corresponding hand positions.
The demonstrations are specific and worth seeing: a wearer plays piano and shoots a mini basketball through a desktop hoop by gesturing, with the robot replicating each motion wirelessly. The same wristband also functions as a computer interface, letting users pinch to resize virtual objects. The hardware currently sits at cell-phone size, and the training data covers a limited range of hand geometries, two constraints the team is actively working to fix.
The paper's real argument is not about the demo tasks. It is about building a large-scale motion dataset to train humanoid robots in high-dexterity work, including surgical procedures. The path from wristband to surgical robot hinges on how well the AI generalizes across body types and movement complexity. Read the full piece to understand exactly where that gap currently stands.
[READ ORIGINAL →]