A key milestone at the brain computer interface has been achieved. A new peer-reviewed neuroscience study led by researchers at the University of California, Los Angeles (UCLA) demonstrates a performance breakthrough in non-invasive brain computer interface (BCI) technology.
“We use shared autonomy to significantly improve the performance of BCIs. Here, Artificial Intelligence (AI) Copilots will work with BCI users to achieve task goals,” San-John Lee, Johannes Lee.
Brain-Machine Interfaces (BMIs) also known as Brain-Machine Interfaces (BMIs) allow users to use their own ideas to control external devices. This can bring life-changing hope to improve paralyzed daily life. The addressable market size for global brain computer interfaces exceeded US$160 billion in 2024, with the non-invasive BCI growth rate for 2025-2023, an estimated combined annual growth rate (CAGR) of 9.35% compared to the 1.49% CAGR of invasive BCI, according to Grand View Research. Examples of BCI providers include Ant Neuro, Blackrock Neurotech, Elon Musk's Neuralink Corp., Synchron, Medtronic, Emotiv, Neurosky, OpenBCI, and other companies.
BCIS can be invasive or non-invasive. Invasive BCIs require neurosurgery and are used to assist patients with epilepsy, Parkinson's disease, cerebral palsy, multiple sclerosis, stroke, spinal sclerosis, amyotrophic lateral sclerosis (ALS) or Lu Gehrig's disease, and motor neuron disease (MND).
Because brain surgery is not necessary, non-invasive brain computer interfaces have broader market potential beyond medical applications, potentially extending to consumers, wearables, gaming, and more. However, non-invasive BCIS has a historically inferior signal-to-noise ratio ratio, resulting in a performance trade-off due to its historically poor signal-to-noise ratio ratio and its accuracy is reduced compared to invasive BCI.
In this study, researchers at UCLA use artificial intelligence to address this performance problem using its pattern recognition capabilities and decipher the noisy data from a non-invasive brain computer interface. The team recorded EEG (EEG) signals from the invasive 64-channel cap in three healthy and one paraplegic participant as they were paralyzed due to T5-level spinal cord injuries and lost motor control and sensation in the legs and lower body. The researchers have created AI algorithms to help decipher brain activity to identify signals that indicate participants' intended movements.
The team used AI convolutional neural network-Kalman filter (CNN-KF). A convolutional neural network (CNN or Convnet) is a deep learning algorithm often used in AI for visual datasets including image or video, natural language processing, speech recognition, and time series analysis. Inspired by human visual systems, convolutional neural networks learn features from data to make predictions.
Imagine trying to estimate past, present and future states with limited knowledge of what you are trying to model. Kalman filters help to solve filtering problems via recursive methods to estimate unknown variables from noisy time series data. It is used in robotics, computer vision, signal processing, time series analysis, aerospace and even air traffic control. The Kalman filter was introduced in 1960 by Hungarian-American Rudolf E. Kalman. asme (American Association of Mechanical Engineers). Think of it as a powerful way to rule out noisy data to find meaningful signals.
“We decode the movement of traditional brain machine interface (BMI) systems from neural data without utilizing target information including the location of potential target locations,” the researchers wrote. “CNN-KF (Convolutional Neural Network Kalman Filter) and Artificial Intelligence (AI) copilots utilize the task structure by updating the decoder parameters of the closed loop (CNN-KF) and modifying the distribution of actions based on observations of the environment in which the user is involved with (copy lot).”
In other words, the team could not only use AI to decode the user's thoughts of the intended action, but also translate the user's intentions and orientation in real time. They created two AI Copilots so that Brain-Computer Interface users can control the robot arm and computer cursor.
The team reported that the AI Copilot solution improved performance by 3.9 times for paralyzed participants in cursor control and robotic arm tasks. Furthermore, researchers report that paralyzed patients were unable to perform the task without the help of AI co-pilot.
This proof of encouragement concept opens the door for the development of more sophisticated AI co-pilots, with more acceleration and accuracy, expanding to more complex tasks in the future.
Copyright©2025 Cami Rosso All Rights Reserved.
