Chinese researchers developed world's first 'dual-loop' non-invasive BCI system, achieving breakthrough in human-machine mutual learning
Chinese researchers from the Tianjin University (TJU) and Tsinghua University have collaborated to develop the world's first "dual-loop" non-invasive brain-computer interface (BCI) system, marking a significant advancement in human-machine mutual learning, the Global Times learned from the Tianjin University on Tuesday.
Researchers told the Global Times that they had successfully employed the system to achieve real-time control of a drone, demonstrating its ability to handle more complex tasks.
The achievement was published in Nature Electronics on Monday.
Practical BCI should be able to decipher brain signals and dynamically adapt to brain fluctuations. This, however, requires a decoder capable of flexible updates with energy-efficient decoding capabilities.
According to the researchers, the latest system is based on a 128k-cell memristor chip. The approach features a hardware-efficient one-step memristor decoding strategy that allows the interface to achieve software-equivalent decoding performance.
The researchers have also developed an interactive update framework that allows the memristor decoder and the changing brain signals to adapt to each other.
Xu Minpeng from the Haihe Laboratory of Brain-Computer Interaction and Human-Machine Integration, TJU, told the Global Times that achieving mutual learning between the brain and machine through information exchange remains a critical and complex challenge in current BCI research.
"Specifically, mutual learning allows machine systems to adjust according to user intentions, enhancing the system's flexibility and autonomy. Users can also continuously optimize their control abilities through interaction with the system, achieving a 'symbiotic' enhancement. This collaborative evolution not only improves the application efficiency of BCI systems, but also drives the overall advancement of intelligent systems," Xu said.
To address the challenge of mutual learning, the joint research team found that the non-stationary characteristics of ctroencephalogram (EEG) signals not only are derived from the background EEG variability as researchers traditionally believed, but also correlate strongly with task-related EEG evolution. Based on this, they innovatively proposed a "dual-loop brain-computer collaborative evolution framework," which was implemented using memristor neuromorphic devices. Under this "dual-loop" framework, the learning loops of the brain and machine collaborate to jointly enhance system performance.
Experimental results demonstrate that this new approach significantly improves decoding speed and reduces energy consumption compared to traditional methods. Notably, in a continuous six-hour interaction experiment, the system's performance not only remained stable, but also achieved an accuracy improvement of about 20 percent. This achievement undoubtedly lays a solid foundation for the practical application of BCI technology, according to Xu.
The performance enhancement can be attributed to two aspects, Xu explained in details. First, this research developed a low-power, high-precision, high-speed brain-computer decoding hardware system: compared to traditional purely digital hardware solutions, the normalization decoding speed of the memristor chip can be improved by more than two orders of magnitude, with energy consumption reduced by less than three orders of magnitude, effectively supporting the successful execution of four-degree-of-freedom drone control tasks. Second, this research realized a new type of BCI system based on neuromorphic memristor devices, where the performance of the BCI did not decline during long-term brain-computer interaction but instead achieved an accuracy improvement of about 20 percent.
When asked for further plan to apply the system in practice, Xu said that that, "on one hand, we need to adopt multimodal brain-computer feedback technology to promote the rapid positive evolution of specific brain functions, exploring the brain's complex, parallel task execution potential to support more complex brain-computer interaction tasks. On the other hand, we need to study the collaborative training model of brain signals and memristor hardware intelligent computing models to achieve more efficient cross-modal information processing and complex decision support."
He predicted that, in the future, the system could could evolve into portable or wearable BCI devices enabling long-term brain-environment interactions and serving as assistive technology in daily life and medical rehabilitation scenarios. It could also be applied in closed-loop neural modulation applications in medical rehabilitation (such as assisting in the treatment of brain diseases or helping stroke patients regain motor functions).