Bimanual Motor-Imagery Brain-Computer Interface Boosts Robotic Rehab Task Success to 90%
Details
Summary
A research team developed a brain-computer interface (BCI) system based on motor imagery (MI) for both hands simultaneously, aimed at training users on robot-assisted bimanual reaching tasks. The system combines machine-learning classifiers that decode the neural signal for intended movement direction with a "shared-control" strategy that corrects decoding errors and manages movement speed through robot autonomy. In a trial with eight participants, task success rate rose from 48.33% with no assistance to 90% at the highest level of robotic assistance.
What happened?
According to a study published in IEEE TNSRE, researchers developed a new online (real-time) BCI system designed for bimanual, robot-assisted training. The system is built on a motor-imagery (MI) model involving imagined coordinated movement in three directions for both hands together: left, center, and right. The system also incorporates a shared-control strategy based on prior knowledge to correct direction-decoding errors, while robot autonomy handles movement speed. The team tested the system across two tasks: a single-step bimanual motor-imagery task and a multi-step, robot-assisted reaching task.
Why it matters?
The study notes that most prior BCI systems designed for robotic rehabilitation focused on motor imagery for a single limb only, while accurately and reliably decoding electroencephalography (EEG) signals remains an ongoing challenge. Real-world daily tasks often require coordination between both hands, making the development of bimanual models a necessary step toward training systems that better reflect real-life needs, rather than being limited to single-limb models.
How the technology works?
The system relies on an array of EEG signals analyzed by various decoding models to identify the intended motor-imagery direction from among three bimanual directions. After decoding, a "shared-control" layer intervenes, combining the user's brain-inferred intent with prior knowledge about the task to correct potential errors in direction detection, while the robot independently adjusts movement speed during execution. This design distributes responsibility between the user and the robot rather than relying entirely on the accuracy of neural decoding alone.
What we know so far
According to the study's findings, six commonly used models were tested to assess the feasibility of the bimanual motor-imagery model in offline decoding trials involving eight participants, and all six models achieved average accuracy notably higher than the chance level of 33.33%. The EEGNet model achieved the highest offline decoding accuracy at 52.93%, compared with a real-time decoding accuracy of 49.67% for the same model. In the real-time multi-step reaching task, the success rate was 48.33% with no robotic assistance, rising to 71.67% with low-level assistance, then to 80.00% with medium-level assistance, and reaching 90.00% with high-level assistance.
What we don't know yet
The trial was limited to eight participants, a small sample that does not allow for broad generalization of the results. The verified facts do not include information on the participation of actual patients (such as stroke survivors), who represent the real target population for rehabilitation, nor on whether performance improvements are sustained over the long term or translate into actual functional benefit. There is also no information available on any direct comparison with single-hand BCI systems in terms of time efficiency or user comfort, nor on any regulatory or commercial pathway for this system, as the work remains at the research stage.
Arab Neurotech's take
The shift from single-limb motor-imagery models to bimanual models represents a logical step toward more accurately simulating natural human movement. In our view, the most important aspect of this study isn't just the improvement in neural decoding accuracy, but the concept of "shared control," which compensates for weaker raw decoding accuracy with intelligent robotic assistance—a practical approach that has proven its worth, given the clear rise in success rates as assistance levels increased. Still, the more important question remains open: how will this system perform with actual patients whose brains have undergone changes due to injury or stroke? That requires broader follow-up studies before any talk of clinical application.
Sources
IEEE Transactions on Neural Systems and Rehabilitation Engineering (IEEE TNSRE): https://ieeexplore.ieee.org/document/11614585
Sources
- IEEE TNSRE Primary source