In a single study published in May 2026 in Cell Reports, researchers at Peking University and Beijing Normal University asked twenty-five volunteers to spend a week learning to fly. The flying happened in virtual reality. Each participant strapped on a headset, looked into a virtual mirror, and saw themselves as a bird-shaped figure with two large, rust-coloured, feathered wings extending from the shoulders. Motion sensors on their arms drove the wings. To lift off, they flapped. To climb, they extended the wings on the downstroke and tucked them on the upstroke. Across four sessions in seven days, they practiced staying aloft, swerving through floating rings, and swatting falling balls out of the air.
Functional MRI scans before and after the training week showed a small but specific shift in the right occipitotemporal cortex, the strip of brain that ordinarily handles the visual recognition of bodies and body parts. After the week, that region responded to images of wings more like the way it responds to images of arms and hands. The shift was not total. The signal for wings still resembled the signal for tools and animal tails as much as it resembled the signal for limbs. But it had moved in the limb direction, and the movement was statistically clear.
The headline version of this result, already circulating in the science press, is that the brain can grow new body parts. That is not what the paper shows, and the authors are careful about saying so. What it shows, in a tightly defined sense, is that one specific visual brain region treats wings somewhat more like body parts after a week of practising with them.
What the experiment was actually built to test
The occipitotemporal cortex, often shortened to OTC, is a piece of brain real estate that has been studied for decades as a kind of perceptual filing cabinet. Different patches of it light up reliably when a person sees faces, places, tools, animals, or body parts. The body-part patch is sometimes called the extrastriate body area, and it shows a consistent preference for images of human limbs and trunks over images of other objects. The standard view, before this study, was that this preference is shaped by evolutionary salience. Humans have always lived in a world with human bodies in it. The visual brain learned to handle that.
The Beijing group, led by first authors Ziyi Xiong and Yiyang Cai, with senior authors Kunlin Wei and Yanchao Bi, wanted to know whether the body-part patch is committed to human bodies in particular, or whether it could expand to include things that look like body parts and behave like body parts, even if they have never existed on any human body anywhere. The wings were chosen specifically because they fit the second category. They look biological. They move in response to the wearer’s own muscles. They serve a clear functional purpose that the wearer is actively pursuing. They are not part of the human form.
The training itself was designed by Cai. According to Science News, the program ran four sessions of about twenty-five minutes each across a week. Tasks included imitating wing postures in front of the virtual mirror, deflecting balls thrown through the air, maintaining altitude over steep simulated cliffs, and flying through hoops. Some participants picked up the controls on the first try. Others took three or four sessions to get airborne. “But you could clearly see them improving,” Xiong told Science News.
What the scans actually showed
The fMRI design was straightforward. Before training, participants in the scanner viewed images of various body parts, various objects, and various wings, while researchers recorded which brain regions responded most strongly to which category. After the week of training, the same procedure was repeated. The comparison the researchers were interested in was whether the response to wings, in body-part-sensitive parts of the OTC, changed in the direction of the response to upper limbs.
It did. According to the paper itself, the right occipitotemporal cortex showed three converging changes after training: increased wing-selective activation in both hemispheres, increased pattern similarity between wings and upper limbs in the right OTC, and strengthened task-dependent connectivity between the right OTC and frontoparietal regions involved in somatosensation and motor planning, specifically when participants were viewing wing images. The paper notes that the coupling change was not observed when participants watched first-person VR flying videos passively, which suggests the effect is tied to active control of the wings rather than to general visual exposure.
The size of the change should be held in context. The pattern for wings shifted toward the pattern for limbs. It did not become the pattern for limbs. The authors note in their discussion that wings, after training, still occupy a kind of in-between status, somewhere between body part, animal feature, and useful tool. That nuance matters, and it tends to get lost in the secondary coverage.
What this is, and what it is not
This is one study. It involves twenty-five healthy adult volunteers, a single laboratory, a single VR system, and a single week of training. The result is interesting, but it should not be read as a settled conclusion about how the brain handles novel body parts in general. The publication is in Cell Reports, which is peer-reviewed, but a single peer-reviewed paper is the start of a research question, not the end of one. Replication in other labs, with larger samples, with different training designs, and with different forms of artificial limb, will determine whether the effect generalises.
The independent reviewer most quoted in the science coverage is Jane Aspell, a cognitive neuroscientist at Anglia Ruskin University in Cambridge. Aspell’s response to Science News was that the study is a nice demonstration of how plastic the brain is, and that if it can incorporate something as unhuman as a wing, it may be able to incorporate other kinds of limb enhancements. That framing, as Aspell offers it, is genuinely cautious. The brain may be more flexible than the textbook account suggested. The next step is to find out how flexible, under what conditions, and for how long.
It is also worth being honest about what the participants did not gain. They did not develop new motor pathways for muscles they do not have. They did not report, in interviews, feeling continuously winged outside the VR sessions. The change documented is at the level of how a specific visual region processes a specific image category, not at the level of overall self-perception.
Why the result is interesting anyway
The reason the paper has attracted attention beyond its narrow technical scope is that it speaks to a question with practical stakes. Prosthetic limbs, exoskeletons, brain-computer interfaces, and other technologies that try to extend the human body all run into a similar problem. The hardware can be built. Its user has to be persuaded, at a neural level, to treat the new hardware as part of the self rather than as an external object the user has to operate. The smoother the integration, the more useful the device.
The Beijing experiment offers a small piece of evidence that this integration can happen quickly, at least at the visual level, and with relatively modest training. A week is a short time in neuroscience. The OTC changes the study documents are not dramatic, but they are clear, and they suggest that the human body schema is willing to extend body-part status to things that look and behave plausibly enough, given enough hands-on practice.
What the experiment does not, and cannot, tell us is what the rest of the system will do. The full sense of embodiment, the feeling that a limb belongs to you, involves a much wider network than the OTC alone. It involves proprioception, motor cortex, parietal integration, and the higher-level sense of self that runs underneath all of them. The paper makes no claims about any of that. Its claim concerns one visual region, in one experimental setup, after one week of structured practice. That claim is interesting in its own right. The rest of the question is still open.
