"Freely Control Robot Arm with Thoughts"… Development of 3D Brain-Machine Interface

Professor Jaeseung Jeong's KAIST Research Team

Professor Jae-Seung Jeong, Department of Bio and Brain Engineering, KAIST

[Asia Economy Reporter Kim Bong-su] A research team led by Professor Jae-Seung Jeong of KAIST, a renowned neuroscientist, has developed a system that allows free control of a robotic arm using only thoughts. This technology is attracting attention as it can be used not only for assistive devices for patients with total paralysis but also in extended virtual worlds (metaverse), smart mobile devices, and gaming equipment.

The Korea Advanced Institute of Science and Technology (KAIST) announced on the 23rd that Professor Jae-Seung Jeong's research team from the Department of Bio and Brain Engineering developed a 'brain-machine interface system' that controls a robotic arm in three-dimensional space with high accuracy (90.9% to 92.6%) using only thoughts. This new type of brain-machine interface system uses artificial intelligence and genetic algorithms to interpret brainwaves measured from the deep cerebral cortex to understand the intention of arm movement and control the robotic arm.

The 'brain-machine interface' technology, which interprets human intentions solely from brain activity and enables robots or machines to act accordingly, has rapidly advanced recently. This technology reads the user's intentions through brain activity and transmits them to robots or machines, serving as a next-generation interface technology applicable not only to robots, drones, and computers but also to smart mobile devices and the metaverse. Previously, commands had to be indirectly delivered through external body parts (buttons, touch, gestures, etc.), but this technology directly transmits commands from the brain, representing the most advanced form.

The challenge lies in accurately interpreting the intention of arm movement direction beyond just the degree of hand movement to precisely control the robotic arm. Brainwaves vary greatly among individuals, and since the system interprets electrical signal characteristics from groups of neurons over a wide area rather than from single neurons, there is significant noise, which limits accuracy.

Brain-Machine Interface Concept Diagram

To address these issues, the research team implemented a cutting-edge artificial intelligence technique called 'reservoir computing' that enables an artificial neural network to automatically learn and identify the critical features of individual brainwave signals required for brain-machine interfaces. In other words, they developed an AI model that recognizes the intention of control 'direction' using only brain activity. It can precisely interpret 24 directions in three-dimensional space with over 90% accuracy. In actual experiments, it decoded 24 directions in 3D space?8 directions per dimension?with an average accuracy exceeding 90% (ranging from 90.9% to 92.6%) across all directions. Additionally, the system successfully moved the robotic arm by interpreting brainwaves generated when imagining moving the arm in 3D space.

Notably, while conventional machine learning techniques such as deep learning require high-spec GPU hardware, this study utilized reservoir computing, allowing AI training on low-spec hardware. This makes it widely applicable to metaverse environments and smart mobile devices.

Professor Jeong stated, "Most brain-machine interface systems that operate robotic arms through brainwaves require high-spec hardware, making real-time applications and integration into smart devices difficult. Our system, with an intention recognition AI accuracy of 90% to 92%, can be extensively used to move avatars in the metaverse as intended or to control apps on smart devices using only thoughts."

The research results were published in the March issue of the international journal Applied Soft Computing.

This content was produced with the assistance of AI translation services.