Brain motor intention Extraction Amplifier: Non-invasive brain-muscle interface
Ye Sun, Bowei Zhao, Dezhong Yao, Rui Zhang, Bohan Zhang, Xiaoyuan Li, Jing Wang, Mingxuan Qu, Gang Liu · 2025
Read Brain motor intention Extraction Amplifier: Non-invasive brain-muscle interface by Ye Sun, Bowei Zhao, Dezhong Yao, Rui Zhang, Bohan Zhang, Xiaoyuan Li, Jing Wang, Mingxuan Qu, Gang Liu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-22895.txt.
Introduction
Ye Sun and colleagues propose routing brain signals through muscle signals to read movement intentions more accurately. Brain-computer interfaces that rely on imagined movement often suffer from sloppy labels, because the recorded brain waves may not match what the user truly intended. Writing in twenty twenty-five, the team builds a non-invasive brain-muscle interface that imitates the natural path from brain to muscle. Muscle electrical signals act as a high-quality relay that sharpens the weak intention signal before it reaches the computer.
Offline tests confirmed the approach works, with prediction accuracy above four fifths in their experiments. In live online tests, every participant succeeded in controlling a virtual arm in a three-dimensional scene. The authors present the relay idea as a way to make motor interfaces more robust for real-world use, where clean labels are hard to guarantee.
Catalog record
| CREATOR | Ye Sun, Bowei Zhao, Dezhong Yao, Rui Zhang, Bohan Zhang, Xiaoyuan Li, Jing Wang, Mingxuan Qu, Gang Liu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 30 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-22895
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Brain motor intention Extraction Amplifier: Non-invasive brain-muscle interface Ye Sun, Bowei Zhao, Dezhong Yao, Rui Zhang, Bohan Zhang, Xiaoyuan Li, Jing Wang, Mingxuan Qu, Gang Liu Abstract Brain-computer interfaces (BCIs) enable real-time interaction between the brain and external devices by decoding neural signals. However, existing motor-based BCI paradigms, like motor imagery BCI, face challenges with imprecise labeling in real-world use. This mismatch between EEG signals and true behavioral intentions leads to pseudo-labels, undermining decoding accuracy and system robustness. To overcome this bottleneck, this paper first proposes a novel motor intention extraction framework based on a non-invasive brain-muscle interface (BMuI)( BCI = Brain Computer Interface = Brain Muscle ( BMuI ) × Muscle Computer Interface \text{BCI}=\frac{\text{Brain}}{\text{Computer}}\text{ Interface}=\frac{\text{Brain}}{{\color[rgb]{1,0,0}\bcancel{\text{Muscle}}}}(\text{BMuI})\times\frac{{\color[rgb]{1,0,0}\bcancel{\text{Muscle}}}}{\text{Computer}}\text{ Interface} ). This method simulates the neural pathway from the brain to the muscles in order to capture and enhance the weak motor intention signals originating in the brain. It then uses EMG as a high-fidelity relay medium to achieve more accurate intention recognition and transmission. To systematically validate the feasibility and effectiveness of this approach, we conducted both offline experiments (to repeatedly verify feasibility) and online experiments (to construct a real-time interactive system and evaluate its performance). The results show that BMuI is feasible, achieving a prediction accuracy of 0.8314; in the online experiment, all participants are able to successfully control the Unity virtual arm. keywords motor intention extraction amplifier, Brain-computer interface,non-invasive brain-muscle interface † † equal-contributors: These authors contributed equally to this work. † † equal-contributors: These authors contributed equally to this work. † † equal-contributors: These authors contributed equally to this work. † † equal-contributors: These authors contributed equally to this work. † † equal-contributors: These authors contributed equally to this work. 1 Introduction BCIs enable direct interaction between the brain and external devices by translating neural intentions into control commands Nicolas-Alonso and Gomez-Gil 2012 ; Wolpaw et al. 2000 ; Santhanam et al. 2006 ; Schalk et al. 2004 . This offers a novel means of communication and environmental control for individuals with severe motor, neurological, or cognitive impairments, significantly enhancing their quality of life Card et al. 2024 ; Metzger et al. 2023 ; Flesher et al. 2021 . Traditional motor-based BCIs, such as motor imagery BCIs (MI-BCIs) Khan et al. 2020 ; Varkuti et al. 2013 ; Ang et al. 2015 ; Cho et al. 2017 ; Roy 2022 ; Khademi et al. 2023 , achieve control by recognising and classifying patterns in EEG. These approaches typically rely on discrete classification labels. In real-world applications, however, the correspondence between EEG and intent labels is often imprecise, resulting in the issue of pseudo-labels, which adversely affects decoding accuracy and system robustness. While such systems are capable of distinguishing between different imagined states to some extent, they are inherently constrained by the low signal-to-noise ratio of EEG and their heavy reliance on discrete labels. As a result, their outputs tend to resemble label-based predictions rather than a fine-grained reconstruction of neural intention strength Maiseli et al. 2023 ; Vilela and Hochberg 2020 , temporal dynamics, or spatial features. Furthermore, the pseudo-label problem exacerbates decoding errors. In complex scenarios–such as multi-degree-of-freedom robotic control, precise prosthetic manipulation, and real-time interaction with the environment–this decoding approach struggles to deliver the high fidelity and multidimensional control required. Consequently, it hinders the deeper application and broader adoption of BCIs in areas such as rehabilitative medicine, intelligent assistive technologies, and human-computer collaboration Bockbrader et al. 2018 . To address the limitations of conventional BCIs-such as pseudo-label noise, limited decoding accuracy, and poor real-time performance–this paper first proposes a non-invasive brain-muscle interface (BMuI). Drawing inspiration from the physiological neural pathway connecting the brain to the muscles AP 1986 , this approach redefines the BCI architecture by explicitly incorporating muscle activity as an intermediate representation. Specifically, it decomposes the conventional brain-computer interface into two functional modules: a brain-muscle interface (EEG to EMG) and a muscle-computer interface (EMG to device). While the latter has been extensively studied and is considered technically mature, the former remains largely unexplored. To bridge this gap, this study presents the first non-invasive BMuI framework that realises continuous EEG-to-EMG decoding with high temporal resolution, effectively filling a critical void in current BCI research. The proposed BMuI establishes a precise, one-to-one temporal alignment between cortical and muscular signals, thereby enabling robust and fine-grained extraction of motor intention. Crucially, EMG not only serves as a high-fidelity carrier of descending motor commands but also functions as a physiological amplifier-transforming weak, noise-prone EEG signals into clearer, more stable representations. By capitalizing on this amplification effect, BMuI enables clearer and more stable decoding of motor intentions, demonstrating marked improvements in transmission accuracy and real-time performance. This provides effective support for achieving natural and continuous human-machine interaction. The contributions of this paper are as follows: 1. This paper presents, for the first time, a motor intention extraction and amplification system–a non-invasive brain-muscle interface (BMuI). Unlike conventional BCIs, which directly map brain signals to external devices, our approach introduces muscle activity as an explicit intermediate variable. Specifically, we decompose the traditional motor-based non-invasive BCI into two sequential modules: a brain-muscle interface (EEG to EMG) and a muscle-computer interface (EMG to device)(see Fig. 1 D). While muscle-computer interfaces are well established and have achieved high accuracy, the brain-to-muscle interface remains largely unexplored. The BMI proposed in this work fills this critical gap by establishing a continuous, high-resolution mapping from EEG to EMG. Both offline and online experiments demonstrate the feasibility and effectiveness of this interface, offering a novel and physiologically grounded pathway for motor intention decoding within neural interface systems. 2. The BMuI proposed in this study achieves, for the first time, the continuous extraction of motor intentions from EEG signals with high temporal resolution. Simultaneously, it establishes a one-to-one correspondence between EEG and the predicted EMG, thereby enabling precise mapping of motor intentions. This groundbreaking development effectively overcomes the limitations associated with pseudo-labels in traditional methods–which often result in reduced accuracy and limited generalisability–and provides a novel technological pathway for motor intention decoding. 3. Although the EMG predicted by the BMuI exhibits significant correlation with experimentally recorded EMG signals, discrepancies remain. To address this issue, the present paper proposes an improved proportional mapping approach: a hybrid control strategy. A convolutional neural network (CNN) is first employed to decode the predicted EMG signals in order to classify the intended movement direction. Subsequently, the muscle with the most accurate prediction and clear physiological relevance is selected for proportional activation mapping. This strategy ensures effective integration with the BMuI framework. Building upon this, we developed, for the first time, an online, real-time neural feedback interface based on the BMuI. This system is capable of accurately decoding EEG, predicting corresponding EMG, and translating the predicted EMG into control commands to drive a Unity-based virtual arm in real time, enabling elbow flexion and extension. The remainder of this paper is structured as follows: Section 2 presents the experimental results; Section 3 discusses the significance of the findings and provides a summary; and Section 4 details the design and implementation of the proposed method. 2 Results Fig. 1 A and B respectively show the online usage and architecture of BMuI, which mimics the neural conduction