Chinese team reveals real-time navigation system in three-dimensional space of the brain to help develop new generation brain-computer interface

China News Service, Beijing, November 13 (Reporter Sun Zifa) How does the navigation system in the brain position the movements of hands and other body parts in real time? This important research direction in the field of neuroscience is crucial to the development of artificial intelligence (AI), especially the research and development and application of brain-computer interface technology, and therefore has attracted much attention.

The Institute of Automation of the Chinese Academy of Sciences announced to the media on November 13 that the team of researcher Yu Shan, director of the Brain Mapping and Brain-like Intelligence Laboratory of the institute, has recently made breakthroughs in this regard. Through primate experimental studies, they revealed the existence of explicit three-dimensional spatial hand position representation in the primary motor cortex (M1, responsible for movement execution) and dorsal premotor cortex (PMd, responsible for movement planning) of the macaque brain.

This study found that the brain not only encodes "how to move", but also maintains a "dynamic three-dimensional hand position map" similar to the Global Positioning System (GPS). In particular, only dozens of position-encoding neurons can accurately decode the three-dimensional hand trajectory, thus providing a solid underlying neuroscience principle for the development of a more efficient new generation of brain-computer interfaces. Relevant results papers were recently published in the international professional academic journal "Journal of Neuroscience".

Division of navigation between two brain regions

According to the research team, in previous studies of the brain's motor cortex, relevant experiments often limited the range of movement of macaques to a small number of discrete positions, resulting in a high correlation between the position of the hand and the direction of movement, making it difficult to distinguish what the neurons are encoding.

To solve this problem, this study designed an unconstrained three-dimensional grasping task. The macaques can completely freely decide when to reach out and where to reach, so that the same spatial position can be repeatedly passed by different trajectories and different directions, thus naturally decoupling "position" and "direction" statistically.

The research team implanted microelectrode arrays in the PMd and M1 on the left side of the macaque to record the discharge of single neurons. By calculating the cell discharge rate during movement in three-dimensional space, they found that 36.2% of the neurons in M1 and 21.3% of the neurons in PMd encoded the real-time position of the opponent. This means that PMd, which is responsible for planning movements, and M1, which issues movement execution instructions, both directly represent the position of the hand in space.

Further analysis showed that the position encoding method of these neurons was not the "linear gradient" as believed by early studies, but appeared as a localized "elongated ellipsoid". The three-dimensional Gaussian model was significantly better than the fitting results of the planar gradient model.

What's more special is that the two "maps" of M1 and PMd are organized in completely different ways: M1's position field is systematically arranged along the main axes of space (front and back, up and down, left and right), and shows the characteristics of multi-scale aggregation (just like a city has blocks and blocks have buildings); PMd's position field direction is more dispersed and the distribution is closer to random.

The research team speculates that this difference may reflect the different division of labor between the two brain areas in movement control. M1 may need to provide a specialized coordinate framework for fine hand control, while PMd is involved in higher-level movement planning and has a more abstract representation structure.

Proving that brain coding is efficient

The research team believes that for brain-computer interfaces, the most direct result of this study comes from the decoding experiment, which uses three groups of neurons: all cells, position cells, and randomly selected non-position cells to decode the three-dimensional trajectory of hand movements.

Experimental data show that in PMd, place cells accounted for only 17.7% of the recording cells, reaching 94.1% of the decoding performance of the entire group; in M1, place cells accounted for 32.8%, reaching 92.9% of the performance. Further analysis showed that to achieve 90% of the decoding performance of the entire population, PMd only needs about 2.8 position-tuned cells and M1 only needs about 16.4. These results confirm the extremely high efficiency of brain coding, and a "few but refined" group of neurons can support high-precision spatial decoding.

At the same time, the research team also found that there is also coding for the grasping target position in the motor cortex of the brain, indicating that the same set of "navigation maps" can not only be used to decode motion trajectories, but also directly decode the position of moving targets. This provides theoretical support for directly decoding motion positions and realizing "what you think is what you get" brain-computer interfaces.

Decoding complex spatiotemporal information

Cao Shenghao, the first author of the paper and a Ph.D. from the Institute of Automation, Chinese Academy of Sciences, said that this study shows that the cerebral cortex is better at macro-intent planning than cumbersome execution processes. If the brain-computer interface can directly extract these planned trajectories, it will not only make the control of external devices smoother, but also greatly reduce the dependence on hardware channels.

Researcher Yu Shan, the corresponding author of the paper, pointed out that the future development of brain-computer interfaces needs to continue to get help from underlying breakthroughs in neuroscience. "The deeper we understand the brain, the better the performance of brain-computer interfaces can be."

It is reported that as one of the series of achievements of Yu Shan's team in the field of spatial coding of the brain's motor cortex, the latest published paper is based on the team's first discovery of hand position field related research in PMd in 2025, extending the research from PMd to M1+PMd, and systematically analyzing the organization of position coding in three-dimensional space, completing a key leap from the motion planning brain area to the planning + execution dual brain area, and from plane position coding to three-dimensional space representation.

The research team revealed that it will further explore the spatio-temporal dimension of PMd in the future, and is expected to upgrade "spatial coding" to "spatial-temporal coding", that is, parse the hand positions in a movement trajectory including the past, present and future. If these complex spatiotemporal information can be decoded, brain-computer interface devices will be able to understand the complete action plan planned by the brain and execute it more efficiently.