Researchers at the Massachusetts Institute of Technology (MIT) have developed an ultrasound bracelet to improve human-robot interaction.
The device, presented in a study led by Xuanhe Zhao, uses high-frequency ultrasound waves to precisely capture movements of muscles, tendons, and ligaments under the skin, and generate detailed data of the human hand useful for training robots in manual dexterity tasks.
How the ultrasound bracelet works
The bracelet incorporates a 256-channel wireless ultrasound system. By emitting and receiving sound waves, it monitors the internal structure of the wrist in real time and records the muscle activity necessary to operate the fingers and palm. According to the researchers, this tracking allows for capturing 22 degrees of freedom of the human hand, that is, its full range of motion for complex and coordinated gestures.
Technology allows for the wireless and remote control of robots and virtual objects. (Screenshot/AP)
More Read
Data processing is performed using a hybrid artificial intelligence model based on Transformer and ResNet architectures. This combination interprets ultrasound signals to replicate muscle activity with greater precision and speed in a portable format.
Advantages over other tracking systems
Precise measurement of hand movements is a historical challenge in robotics and spatial computing. Camera-based systems have limitations due to viewing angles and obstacles. Other solutions, such as bands with strain sensors, inertial sensors, or electromyography, often restrict the range of motion or do not offer the resolution necessary to capture continuous finger gestures, as detailed by MIT.
The ultrasound bracelet avoids those problems by allowing continuous and wireless capture of the hand’s internal activity, without interfering with the user’s natural movement.
More Read
During testing, the bracelet recognized all 26 letters of the American Sign Language with high accuracy. (Screenshot/AP)
Laboratory results and demonstrated applications
In the tests, eight volunteers used the bracelet to reproduce hand gestures, with a tracking fidelity of 120 milliseconds of latency. The system recognized the 26 letters of the American Sign Language alphabet, which showed its ability to capture complex finger and palm configurations.
Technology allows for controlling robots remotely, manipulating three-dimensional objects in virtual reality environments, and guiding a robotic hand to execute complex actions, such as playing the piano.
The results were published in the journal ‘Nature’.
A tool for the next generation of humanoid robots
Beyond the remote control, the MIT team envisions the bracelet as a way to create large databases of human hand movements. Those datasets could train future humanoid robots and accelerate the autonomous learning of sophisticated manipulation skills.
The initiative seeks to create massive databases of manual movements to train humanoid robots.
The proposal joins other advances in wearable technology and ultrasound, such as devices for continuous monitoring of internal organs, cardiac imaging, or cerebral flow. The goal is to bring robotic dexterity to levels comparable to those of the human hand and expand applications in automation, remote assistance, and human-machine interaction.
