BCI Controlled Robot Contest on the 50th Anniversary of Brain-Computer Interfaces

Introduction

Is it feasible to employ neural signals for the purpose of communication? In 1938, this thought- provoking query was visually embodied in a sketch received by Hans Berger, a German Professor of Psychiatry renowned for his discovery of human electroencephalography (EEG) nearly a century ago. In this fanciful sketch, a Christmas greeting from a young Canadian neuroscientist Herbert Jasper, i.e., “I am wishing you a pleasant Yuletide and a New Year as you like it”, can be vividly translated from signals recorded from the human brain, and this is an early conceptualization of brain-computer interface (BCI). The scientific term of BCI was not coined until 1973, when Jacques J. Vidal, a professor at the University of California, Los Angeles, brought the term of BCI into literature and introduced a technical framework for human-computer communication via BCIs. Building upon his conceptual framework, Vidal systematically discussed the feasibility of interactive control in computer tasks via brain messages identified from single-trial ongoing EEG and in his subsequent work, he implemented a BCI system based on visual evoked potentials (VEP) in which subjects can perform an online maze navigation task by making selections from four targets. Professor Vidal’s pioneering work established BCI as a novel discipline in academic literature, a seminal milestone that we commemorate on its 50th anniversary this year.

The technology of BCI, also known as the brain-machine interface (BMI), establishes a direct conduit to extend the human brain to external devices, such as computers or actuators, augmenting human sensory and motor functions. Over the past 50 years, generations of neuroscientists and engineers have devoted continuous effort to enriching and refining the gamut of BCI research, pushing the boundary of BCI toward real-life applications. For the BCI research, substantial progress has been made in both non-invasive and invasive BCIs, in which during the past 15 years the former have witnessed a surge of information transfer rate and the latter have garnered a growing excitement about moving from bench to bedside. For the application side, BCI competitions offer a rare opportunity to promote and advance BCI toward practicality in a way that brings together diverse state-of-the-art BCI paradigms, applications and decoding techniques in one arena. As an integral part of the World Robot Contest annually held in Beijing, the BCI Controlled Robot Contest, organized by Tsinghua University and the Chinese Institute of Electronics with a joint effort of 9 institutional co-organizers and 14 institutional partners, serves as a top-notch national platform for BCI competition in China and a unique opportunity for researchers to showcase their state of the art in BCI technology.

In 2022, the BCI Controlled Robot Contest attracted more than 100 top universities, research institutes and enterprises in the BCI community, and the number of participants reached over 3000. Analogous to the competition in the previous year, this year’s contest comprised four principal categories: algorithm competition, project contest, youth outstanding paper award, and project exhibition. Specifically, the algorithm competition intended to identify and award the cutting-edge algorithms, while the purpose of project contest was to single out the subjects with exceptional BCI performance. In the algorithm competition, five BCI tracks were provided for contestants, including steady-state visual evoked potential based BCI (SSVEP-BCI), motor imagery BCI (MI-BCI), rapid serial visual presentation based BCI (RSVP-BCI), affective BCI and BCI Turing test. The BCI competition for each track comprised two stages, namely the preliminary and final rounds, which corresponded to offline BCI tasks and online BCI tasks, respectively. In the final rounds, winners of the project contest were selected as the subjects of the online experiment, with each subject completing all five BCI tracks in a row.

The Special Issue on BCI Competition and Selected Algorithms is initiated to foster the research and expedite the application of outstanding projects from the BCI Controlled Robot Contest in World Robot Contest. The Volume 9 Issue 3 of Brain Science Advances, together with its preceding [1] and forthcoming editions, is part of an ongoing effort toward this goal. As an outcome of BCI Controlled Robot Contest in 2022, the collection of special issue this year is composed of seven articles, including five papers corresponding to the main BCI tracks, a paper on the MATLAB undergraduate group and an additional paper on the adolescent project contest. In this issue, we highlight emerging trends in BCI decoding methods, particularly the increasing confluence of artificial intelligence (AI) and human intelligence (HI). This convergence is evidenced by the evolving decoding methodologies poised to address intricate challenges, notably affective BCI and the BCI Turing test.

The subsequent section provides a brief outline corresponding to each article included in the issue:

Du et al. [2] reviewed the asynchronous decoding methodologies employed by the leading five teams in the SSVEP-BCI track, comparing an array of canonical correlation analysis (CCA) based dynamic stopping strategies for distinguishing between intentional control (IC) and non-control (NC) states in asynchronous SSVEP-BCI decoding.

An et al. [3] conducted a comprehensive survey on existing decoding approaches for MI-BCI in the literature, and further provided a detailed comparison of EEGNet based backbone models implemented by the top five teams of the MI-BCI track in terms of channel selection, data length selection, data preprocessing, data augmentation, classification network, training, and testing settings.

Wang et al. [4] performed a comparative evaluation of the top five approaches used by various teams in the RSVP-BCI track, in which deep learning based models were widely adopted by the teams to tackle the cross-subject event-related potential (ERP) decoding and an in-depth analysis of feature maps underlying the neural network was presented.

Yi et al. [5] summarized the paradigm and competition rules of the newly introduced BCI Turing Test, and presented an overview of the winning algorithms applied to decode user intent from a hybrid SSVEP-BCI and motor imagery task, suggesting the exciting possibilities in using BCI to discern whether control commands are issued by human intelligence or artificial intelligence in human-computer interactions.

Si et al. [6] proposed a Transformer-based ensemble (TBEM) model that comprised a pure convolutional neural network (CNN) and a cascaded CNN-Transformer hybrid model for EEG-based emotion recognition. By leveraging ensemble learning and re-referencing preprocessing, the proposed TBEM method was able to effectively discriminate between eight emotional states and outperform other competing methods in the affective BCI track.

Yi et al. [7] conducted a detailed comparative analysis of three SSVEP-BCI decoding methods employed in the MATLAB undergraduate group and found that online adaptation CCA (OACCA) + dynamical window (DW) achieved the highest performance, followed by filter bank CCA (FBCCA) + DW and then spatiotemporal equalization (STE) + DW, demonstrating the efficacy of the BCI Controlled Robot Contest in cultivating the novice practitioners in the BCI domain.

Dong and Tian [8] proposed a large database towards user-friendly SSVEP-BCI in the adolescent project contest, in which 59 healthy volunteers participated in the 40-target SSVEP speller using ergonomically designed semi-dry electrodes and grid stimuli, underscoring the potential for deploying robust and efficient SSVEP-BCI systems in practical applications.

In summary, 50 years after the term BCI debuted in the literature, the convergence of a broad set of disciplines at an ever-increasing pace is bringing the field of BCI into full bloom. At this juncture, the BCI Controlled Robot Contest will continue to serve as the melting pot to hasten innovative breakthroughs in BCI, nurture and support the next-generation scientists and engineers in the BCI community over the coming 50 years.