Currently, invasive brain-computer interfaces (BCIs) in China have reached 4,096 channels, but fundamental research in neuroscience remains weak. Elon Musk once strongly advocated for the invasive route, with the core idea of parsing and digitizing the brain. However, the real-world indications currently addressed by invasive BCIs are merely initial trials, and their scientific value has not yet been fully realized.
On the afternoon of November 13, at the 5th Nandujiang Smart Healthcare and Rehabilitation Industry Summit Forum, Li Luya, Founder and Chairman of Angelexo Science and Technology (Hangzhou) Co., Ltd., shared his views on the implementation paths of BCIs regarding how BCIs, AI, and robotics reconstruct the underlying logic of neural intelligence.
Li stated that BCIs have now become a national policy. Commercial companies, standing at the forefront of clinical practice and product application, are more concerned with whether they can read brain signals, understand them, form closed loops, and solve problems in clinical settings and daily life.
He believes that while non-invasive devices are convenient—"you can even buy an EEG cap on Taobao"—they cause confusion when used on the front lines. The biggest headache is motion artifacts. While sitting still is fine, once movement is incorporated into rehabilitation, the signal under rigid connections is severely compromised, decoding rates are suppressed, and this type of BCI hits its ceiling.
Angelexo is pursuing two paths. One focuses on stability, using flexible electrodes for "use-as-you-collect," tolerating signal changes during movement. The other falls into a category without formal classification yet—"minimally invasive non-invasive": electrodes penetrate the stratum corneum and enter the dermis, can be worn at the same location for at least two weeks, ensure safety, and allow bipolar placement. These electrodes require no protection; once worn, the sensors remain stationary, addressing the bottleneck of "scalable usage."
At the conference, Li demonstrated the effectiveness of two forms: flexible velcro electrodes already performing well clinically, and microneedle electrodes inserted into the scalp via a microneedle gun, reaching about 5mm below the stratum corneum. The stratum corneum is the biggest barrier; once breached, impedance drops below 5K, and dual channels can form a closed loop.
After reading comes understanding. Does the patient want to move their arm or leg? More troublesome is that a stroke patient's brain signals differ completely between day one and day thirty of rehabilitation. Li believes AI can handle this: algorithms adapt to patients, data is retained after one wear, and automatic adjustments occur throughout the cycle. Combined with the microneedle structure, the decoding ceiling can rise from 80% to 90%, adaptation for different individuals and times occurs within minutes, and real-time control precision is compressed to within 100 milliseconds.
As for who judges "how well we read," he proposed relying on brain imaging—using clinical outcomes to validate effectiveness and verify if the research path is correct.
Beyond reading and understanding, writing to the brain is also critical, arguably the watershed moment for BCIs. There are many methods for reading the brain—three main routes plus various variants—but few methods exist for writing to it. Writing can utilize light, electricity, or sound. Li advocates for simpler means, leaving the rest to be refined through clinical practice.
He cited depression as an example: previous treatments lacked closed loops, and dosages were determined by trial and error—should we apply one milliampere or two? Thirty minutes or twenty? How do we define a cure? "Now that BCIs enable closed loops, problems that previously could not be solved precisely can now be addressed accurately." By regulating nerves through excitation and inhibition, neuronal activity around lesions can be enhanced for specific rehabilitation stages, and hunger sensation can even be inhibited when depth is sufficient. Future efforts will advance according to different indications.
Once reading, understanding, and writing are resolved, the next step is integration with robots—actions performed by robots, with effects fed back into treatment plans. Li noted that Angelexo currently has products covering shoulder, wrist, and hand levels, and has conducted extensive brain imaging to verify brain recovery status. Based on this, the company is building a platform that opens up the acquisition, decoding, and execution layers.
Li judges that BCIs are moving from single-signal acquisition to bidirectional controllability, and even toward dedicated large model platforms. However, from a clinical perspective, for technology to truly transform from national-level strategies, systems, and policies into tools serving the people, scenarios must be well-defined, combining medical insurance coverage with consumer willingness.
"For the next three to five years, my personal view is to solidify clinical applications, obtain registrations, and finally expand through channel promotion," he said. The concept of BCIs has already blossomed in the market, with each player having independent indications, and the core orientation being service to the people and to all humanity.