A person flew a drone using nothing but their thoughts, and the device kept getting better at reading them. Researchers in China built what they call the first two-way brain-computer interface, where the brain and the machine adjust to each other instead of the machine simply reading signals. The work was published in the journal Nature Electronics on 17 February 2025.
A brain-computer interface, or BCI, links brain activity to a device so a person can act by thinking. Making that link work in both directions could make these tools steadier over time and useful in far more settings, from medical care to everyday control.
A brain-computer link that talks back
Traditional BCIs are one-way. They watch brain waves, learn to recognize patterns, and turn them into commands, such as moving a cursor. The new system from Tsinghua University and Tianjin University adds a return path, so the device also sends feedback to the brain through a dual-loop design. One loop updates the decoder as brain signals drift, and the other helps the user sharpen their own control.
That second channel matters because brain signals are not fixed. They shift partly in response to the device itself, which can make a one-way decoder less reliable as time passes.
When the brain and the chip learn together
At the core of the system is a decoder built on a 128,000-cell memristor chip, a type of brain-inspired hardware that stores and processes information in the same place. A one-step decoding method let this chip reach performance on par with a software decoder running on a normal computer.
The headline idea is co-evolution: the decoder and the brain adapt to each other over repeated use. Across an extended task with ten participants, this co-evolution gave around 20% higher accuracy than a version without it. The choice of memristor hardware fits a wider push toward brain-like chips that cut energy use while handling neural signals.
Flying a drone with brain signals
To show the system in action, the team used it for hands-free control of a drone. Brain signals alone steered the drone in four degrees of freedom, including rotation and forward or backward motion, in real time.
Richer control like this is the practical payoff of a two-way link: by recognizing more patterns, the interface lets a user carry out more complex tasks than a simple on-off command.
Efficiency gains and what is still unproven
The team also reports large efficiency gains. Compared with conventional BCIs, they say the device boosts efficiency about 100-fold and cuts energy demand roughly 1,000 times, which is the kind of saving that could let a BCI run on portable and wearable devices for consumer and medical use.
These numbers come from the researchers’ own benchmarks, not independent testing, and the study involved only ten people in controlled lab conditions. The medical goals, such as helping people with brain damage regain lost abilities, remain future targets rather than proven results. Progress here sits alongside other national efforts to build brain chips for direct brain-device communication.
Sources and related information
Nature Electronics – A memristor-based adaptive neuromorphic decoder for brain-computer interfaces – 2025
The peer-reviewed paper reports a decoder on a 128,000-cell memristor chip that co-evolves with the brain, raising accuracy by about 20% over ten participants and controlling a drone in four degrees of freedom.
TechXplore – First two-way adaptive brain-computer interface enhances communication efficiency – 2025
Bob Yirka’s report explains the dual-loop design and the reported 100-fold efficiency gain, and how the added feedback path lets users perform more complex tasks.
South China Morning Post – Chinese scientists make brain-computer co-evolution possible for the first time – 2025
The report frames the work as a first step toward mutual adaptation between brain and machine, aimed at portable and wearable devices for consumer and medical use.
