Related Experiment Video
Updated: Aug 11, 2026
10:14
Motor Imagery Performance Through Embodied Digital Twins in a Virtual Reality-Enabled Brain-Computer Interface Environment
Published on: May 10, 2024
Hands-free motor imagery EEG classification via LLM multi-agents
Ruiyu Zhao1, Shurui Li2, Xinjie He1
1The Key Laboratory of Smart Manufacturing in Energy Chemical Process, Ministry of Education, East China University of Science and Technology, Shanghai, 200237, China.
Journal of Neuroscience Methods
|August 4, 2026
Summary
AutoMI automates the design of motor imagery (MI) brain-computer interface (BCI) models using multi-agent iterations. This novel framework significantly enhances MI-EEG classification accuracy and efficiency.
Area of Science:
- Neuroscience
- Computer Science
- Biomedical Engineering
Background:
- Motor imagery (MI) brain-computer interfaces (BCI) depend on accurate electroencephalogram (EEG) classification.
- Current MI-EEG models face challenges in manual design, parameter tuning, and task flexibility, limiting state-of-the-art (SOTA) efficiency.
Purpose of the Study:
- To introduce AutoMI, a novel framework for automated construction of SOTA MI-EEG models.
- To overcome limitations of manual design and improve task flexibility in MI-EEG models.
Main Methods:
- AutoMI employs multi-agent automated rapid iterations for constructing MI-EEG models.
- A hybrid decision mechanism combines Q-learning with deterministic rules.
- Integrated planning, execution, and output agents with tools, plus experience tracking and rollback mechanisms, ensure broad applicability and prevent optimization ambiguity.
Main Results:
- AutoMI-constructed SOTA models achieved accuracies of 77.62% (IV2a), 78.08% (OpenBMI), and 83.02% (ECUST-MI).
- Maximum accuracy improvements reached 24.69%, 23.35%, and 23.28% respectively.
- Compared to other automated algorithms, AutoMI showed accuracy increases of 18.42%, 9.27%, and 19.25%.
Conclusions:
- AutoMI demonstrates superior optimization capability, reaching SOTA performance in MI-EEG classification.
- The framework offers a novel perspective and design reference for future BCI model optimization.
