A student puts on a soft EEG headband, thinks about moving a drone forward, and the rotorcraft eases ahead — then turns, climbs and rotates, all from continuous intention rather than a single click. No surgery, no implant, just a signal most labs once considered too faint to use.
That is the kind of interface Xu Minpeng (许敏鹏) has spent his career making work.
Key takeaways
- Xu Minpeng is a brain–computer interface researcher at Tianjin University and a member of the university's Brain–Computer Interface Haihe Laboratory (脑机海河实验室).
- As a PhD student he was first author of an IEEE TBME cover article on a brain speller that decodes EEG signals as weak as about 0.5 microvolts, under corresponding author Ming Dong (明东).
- He co-led a "dual-loop" (双环路) memristor-based, two-way adaptive BCI with Ming Dong and Tsinghua University, published in Nature Electronics.
- The system gives a user four degrees of freedom of continuous drone control and, over six-hour sessions, improved accuracy by roughly 20% instead of decaying.
- The lab frames the work as the world's first non-invasive adaptive BCI of its kind, aimed at wearables, rehab and hands-free control.
A researcher who stayed with the weak signal
Xu earned his PhD in biomedical engineering at Tianjin University in 2015 after a visiting stint in computational neuroscience at the University of California, San Diego. His early signature result was unglamorous: finding order in noise. In an IEEE Transactions on Biomedical Engineering cover article, his team showed a speller that could read the smallest visual-evoked potentials — brain signals around 0.5 microvolts, far below the 2-microvolt floor most systems needed. The point was not a party trick but a principle: non-invasive BCI's limit is signal quality, and signal quality is an engineering problem.
That principle now anchors a larger programme at the BCI Haihe Laboratory, where Xu works alongside Ming Dong on interfaces that do not require opening the skull.
The link that learns back
Traditional BCIs are one-way: the brain speaks, the machine decodes, performance drifts as the user tires. Xu's group attacked the drift with a "dual-loop" design built on a memristor neuromorphic chip. One loop updates the decoder as the brain signal shifts; the other sends feedback so the user learns to control the machine better. Over a six-hour session the two adapt together, and accuracy climbed by about 20% rather than falling — the opposite of the usual fatigue curve.
The work, done with Tsinghua's integrated-circuit team (Wu Huaqiang and Tang Jianshi), was published in Nature Electronics and demonstrated as four-degree-of-freedom control of a drone: forward–back, left–right, up–down and rotation, all from continuous thought. The lab describes it as the first non-invasive adaptive system of its kind. The promise is practical — firefighters or inspectors with both hands busy, or patients steering assistive devices — because it needs no surgery and can in principle shrink to a wearable. The lab is explicit that the target is not a one-off demo but a category: surgery-free control for rehabilitation, assistive robotics and industrial inspection, where implant risk has kept BCIs off the shop floor.
Why non-invasive is the Chinese front
China's BCI policy explicitly favours non-invasive scale: Ming Dong, Xu's senior colleague, has argued that high-performance non-invasive interfaces reaching ordinary users — not everyone receiving an implant — are what will build the industry. Xu's co-adaptive approach is a direct answer to the field's hardest objection: that scalp signals are too noisy and too unstable for real use. His group's answer is that noise is trainable. For China, where the 15th Five-Year Plan names brain–computer interface (脑机接口) a future industry, non-invasive scale is also a policy preference, not just an engineering one.
According to Li An, chief scientist at BrainNet (脑机网), non-invasive co-adaptive interfaces are where Chinese teams are currently most competitive, because the bottleneck is engineering and data density rather than surgical risk or implant durability.
Honest limitations
The performance gains are reported by the research team and a handful of outlets covering the Nature Electronics paper; the ~20% six-hour improvement is the team's measured result, not yet an independently reproduced benchmark across labs. Non-invasive BCI still trails invasive recording on signal richness, and long-term durability, daily-use reliability and clinical rehab efficacy at scale remain unproven. The "world's first" claim is the laboratory's own framing. The work is a research demonstration, not a shipped product, so timelines to any wearable or medical use are uncertain.
What readers can do now
- Clinicians and researchers: read the Nature Electronics paper and the earlier IEEE TBME speller work as the two anchors of Tianjin's non-invasive programme.
- Patients and families: track non-invasive rehab and assistive trials from the BCI Haihe Laboratory rather than assuming implants are the only path.
- Investors: watch the lab's commercialization partners for wearables and hands-free control, where surgery-free BCI has the widest addressable market.