MIT researchers use AI to convert hand gestures into data for training robots

design sem nome (15)
Humanoid robots learning tasks like grasping a cup now have a new instructor: a person wearing an ultrasound wristband that records the movement of muscles, tendons, and ligaments beneath the skin.
Xuanhe Zhao, MIT mechanical engineering professor, poses with an ultrasound wristband that enables a robotic hand to mimic full hand motions in Cambridge, Mass., on April 14, 2026.
Image Credits: AP Photo/Rodrique Ngowi

Humanoid robots learning tasks like grasping a cup now have a new instructor: a person wearing an ultrasound wristband that records the movement of muscles, tendons, and ligaments beneath the skin.

MIT System Improves Robotic Dexterity

Researchers at the Massachusetts Institute of Technology (MIT) created a system to capture human hand movements, with the goal of helping robots develop the fine motor skills that remain difficult for machines to replicate.

“Imagine people doing housework,” said Xuanhe Zhao, an MIT mechanical engineering professor. We can use our system’s data to train robots to perform the same tasks with similar dexterity.

While much of the tech industry focuses on AI assistants for digital tasks, Zhao and his team aim to give AI richer real-world sensory data.

Wristband Replicates Hand Movements Using AI

The technology could extend beyond household chores to other activities that demand precise finger and hand movements, including surgical procedures.

The wristband works by using high-frequency sound waves to “look” beneath the skin, capturing images of muscle and tendon activity. The signals are sent to a computer, where AI processes them and controls a robotic hand to replicate the user’s gestures.

Mit Researchers Channe 1
Massachusetts Institute of Technology mechanical engineering graduate student Dian Li demonstrates how an ultrasound wristband can help a robotic hand mimic full hand motions in Cambridge, Mass, on Tuesday, April 14, 2026. Image Credits: AP Photo/Rodrique Ngowi

An AI model converts device images into “degrees of freedom,” the ways joints bend and rotate; the human hand has 22. Earlier systems struggled to accurately capture even a small portion of such movements.

Wristband Reproduces ASL Gestures in 120 Milliseconds

In tests with eight volunteers, the wristband accurately replicated hand gestures, including all 26 American Sign Language letters, in about 120 milliseconds.

The wristband can function wirelessly, so the user and the robotic system do not need to be in the same location.

Beyond enabling remote operation, the researchers believe it could be used to collect large-scale datasets of human movement, which may ultimately allow humanoid robots to learn complex dexterous skills without direct human instruction.

WhatsApp Image 2026 03 21 At 15.37.18 1 768x384 21

Read the original article on: Tech Xplore

Read more: Simple Color Indicator Makes Prosthetic Devices Easier to Master

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top