Editorial: Long Term User Training and Preparation to Succeed in a Closed-Loop BCI Competition

Gernot Müller-Putz, Damien Coyle, Fabien Lotte, Jing Jin, David SteyrlFrontiers in Human Neuroscience 2022 · DOI · OpenAlex

Introduction: Bridging the Gap Between Lab and Life

For over three decades, Brain-Computer Interfaces (BCIs) have promised a revolution in independence for individuals with severe motor disabilities. However, a persistent "lab-to-life" gap remains. While BCI systems often perform admirably in controlled academic settings, they frequently struggle when deployed in the complex, noisy, and unpredictable environments of the real world. This editorial explores how high-stakes competitions, such as the CYBATHLON and the China BCI Competition, serve as critical catalysts for moving BCI technology out of the laboratory and into practical, end-user applications.

The Competition as a Stress Test

Competitions like the CYBATHLON, initiated by ETH Zurich, challenge research teams to demonstrate their systems in real-world scenarios. In the BCI Race, "pilots" (end-users with disabilities) control digital avatars using multi-class BCIs. These events are uniquely demanding: the system must function flawlessly at a specific time, in a foreign environment, amidst spectator noise, and under significant psychological pressure.

Similarly, the China BCI Competition, organized by the China Electronics Society, involves thousands of participants annually. This event consists of both a user competition and an algorithm competition, generating massive datasets that help benchmark and refine BCI algorithms. These "stress tests" force developers to move beyond static datasets and address the "closed-loop" reality: a system where both the machine and the human user must learn to adapt to one another simultaneously under pressure.

Methodological Advances in Signal Processing

To succeed in these environments, researchers are moving beyond static classification models. The editorial highlights several key methodological shifts synthesized from recent studies:

  • Adaptive Signal Processing: Utilizing Riemannian classifiers and advanced feature selection methods (analyzing multiple time segments during a task) allows systems to handle the inherent instability of brain signals—known as signal drift—over time.
  • Transfer Learning: By implementing inter-session transfer learning, teams can significantly reduce the grueling calibration time typically required before a BCI can be used, making the technology more accessible.
  • Closed-Loop Calibration: Traditional "open-loop" calibration, where a user passively watches a screen, is being replaced by closed-loop systems. These provide real-time feedback, which has been shown to be more engaging and physiologically effective in eliciting localized and distinct brain activation patterns.

The Human Element: Training and Co-Learning

Perhaps the most critical factor in BCI success is the preparation of the pilot. Research indicates that BCI control is a "co-learning" process where both the machine and the human must adapt.

  • Long-Term Training: Longitudinal studies, such as the 14-month training protocol described by Hehenberger et al., demonstrate that consistent practice leads to more stable and accurate mental states. This training helps pilots produce consistent brain oscillations even in stressful race conditions.
  • The Impact of Emotion: BCI performance is not just about signal-to-noise ratios; it is inextricably linked to the user's psychological state. Studies show that a pilot's emotional state significantly impacts their ability to control complex 4-class BCIs, suggesting that future systems must account for physiological markers of stress and performance anxiety.
  • Neural Adaptation: In a striking example of co-learning, research shows that long-term training can lead to a phenomenon where the user’s brain patterns shift to more closely resemble the EEG patterns the classifier was originally trained to recognize—the human literally meets the machine halfway.

Key Results and Future Directions

The synthesis of these studies reveals that success in closed-loop BCI applications requires a dual focus on algorithmic adaptability and rigorous user training. Key results from the featured research include:

  • Enhanced Stability: Long-term training significantly improves performance in both calibration and competitive "game" scenarios.
  • Algorithmic Robustness: Adaptive Riemannian classifiers have proven successful in addressing both day-to-day (inter-session) and within-session (intra-session) signal variability.
  • Data-Driven Progress: The large-scale datasets generated by competitions like the China BCI Competition are becoming vital resources for the global research community to benchmark new algorithms.

Looking forward, the authors acknowledge that the optimal integration of machine learning and human learning remains an open question. Future work must focus on standardizing user-centered training protocols and developing BCIs that are inherently robust against the physiological effects of stress. The ultimate goal remains clear: transitioning BCIs from experimental curiosities into reliable tools that enhance the quality of life and independence of their intended users.

Summary updated

Abstract

Computer Interfaces (BCIs) have been researched now for about 30 years. However, even<br/>still,much of the work done in the lab environment is seldomapplied within the targeted end-users,<br/>for example, people with severe motor disabilities.<br/>The main goal of the research community should be to finally bring the BCI into a state in<br/>which the end-users can profit and gain independence and quality of life. One possibility to push<br/>the field into practical applications is driven by initiatives such as CYBATHLON [initiated by ETH<br/>Zurich (Riener, 2016)] and other competitions. Such a competition challenges research institutions<br/>and industry to demonstrate their developments in real-world scenarios and push the boundaries<br/>of research.<br/>In the BCI Race, at CYBATHLON (Novak et al., 2017), end-users are the pilots and they control<br/>an avatar in a race against other pilots, by using a multi-class BCI. This competition and others are<br/>enormously demanding to the developers, as the BCI system must work properly at the time of the<br/>competition, out of the lab in a foreign environment, with spectators around, noise and without<br/>second chances.<br/>In China, a BCI competition was firstly organized by Tsinghua University in 2010. Since 2017,<br/>the BCI competition has been organized by China Electronics Society as part of World Robotics<br/>Conference. Thousands of users participate every year. This BCI competition consists of two parts,<br/>a user competition and an algorithm competition. Winners of the user competition compete then<br/>in the algorithm competition to test the performance of algorithms which were uploaded by BCI<br/>research teams. Through these BCI competitions, a large amount of BCI data for further research<br/>has been obtained, which has been used to promote the progress of BCI algorithms. In the near<br/>future, these data will be published online for BCI researchers all over the world.<br/>Another extremely important factor, of course, is the preparation of the team for the<br/>competition. In particular, the end-user pilot should be trained to produce stable and accurate<br/>mental states that produce consistent brain oscillations to control the BCI, even in a potentially<br/>stressful environment such as in the CYBATHLON race arena.
BibTeX
@inproceedings{mullerputz2022editorial,
  author = {Gernot Müller-Putz and Damien Coyle and Fabien Lotte and Jing Jin and David Steyrl},
  title = {{Editorial: Long Term User Training and Preparation to Succeed in a Closed-Loop BCI Competition}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2022},
  doi = {10.3389/fnhum.2022.869700},
}
Figure