Physical AI News (Japan)

Physical AI, robotics and compute — reported from Japan.

日本語
· Research & technology· 日本語で読む

HKUST Develops Artificial Eye Combining Full-Color Imaging and Event-Driven Motion Detection

The Hong Kong University of Science and Technology has developed an artificial eye with a hemispherical artificial retina that combines full-color imaging and motion detection. It has a pixel density of 1,905 ppi and 367,500 pixels, with an aberration-corrected field of view exceeding 160 degrees. Under the study’s test conditions, it reduced data transmission by more than 99.95% and achieved 98.6% motion-recognition accuracy compared with conventional frame-based imaging.

Key facts

  • The hemispherical artificial retina has a pixel density of 1,905 ppi and a total of 367,500 pixels.
  • Its field of view exceeds 160 degrees and supports aberration correction.
  • The photosensitive layers are sensitive to blue, green and red light, with a spectral response range of 300–800 nm.
  • Under the study’s test conditions, it reduced data transmission by more than 99.95% and showed 98.6% motion-recognition accuracy compared with conventional frame-based imaging.
  • The HKUST team developed a three-dimensional laser-processing method to form high-density pixels on a hemispherical substrate.

What happened

The Hong Kong University of Science and Technology (HKUST) announced on November 13, 2026, a hemispherical artificial eye that integrates full-color imaging, an aberration-corrected field of view exceeding 160 degrees and event-driven motion detection. The artificial retina has a pixel density of 1,905 ppi and a total of 367,500 pixels. Under the study’s test conditions, it reduced the data transmission required for motion recognition by more than 99.95% compared with conventional frame-based imaging and achieved 98.6% recognition accuracy.

Why it matters

HKUST said the artificial retina not only captures images but also directly detects dynamic changes, reducing data transmission by more than 99.95% compared with conventional frame-based methods under the test conditions. It is a concrete example of technology being evaluated to determine how far a configuration in which the sensor handles motion detection can be applied to robots and autonomous systems, where visual data transmission loads are a challenge.

Analysis

The focus of this research is not simply increasing the number of pixels in an artificial eye, but combining image capture and motion detection on a hemispherical sensor. HKUST said three-dimensional laser processing was used to form high-density pixels across a hemispherical surface, enabling the same artificial retina to provide full-color imaging and event-driven motion detection. The research presents the processing method and the scale of the pixel array together as a response to the manufacturing challenges of curved sensors. HKUST said conventional frame-based methods process entire successive images to extract motion, while this approach directly detects dynamic changes in images. This provides a basis for reconsidering how image-capture sensors and motion-extraction processing are divided in robotic vision systems. However, the reported reduction of more than 99.95% refers to data transmission under the study’s test conditions; it does not mean power consumption or the total computational load of the system fell by the same percentage. Nor does the 98.6% recognition accuracy directly guarantee performance in other environments or tasks. The integration of blue-, green- and red-sensitive layers, an aberration-corrected field of view exceeding 160 degrees and event-driven detection in one hemispherical structure also raises questions about vision-module design, including optics and downstream processing. Although the system provides a wide field of view and color images, the announcement does not establish that it also provides the depth information or object identification required by robots. Further testing is needed to determine how much of existing recognition systems it can replace. The next areas to watch are the manufacturing reproducibility of the pixel array using three-dimensional laser processing, image quality and motion detection in real-world settings, sensor power consumption, and integration into robots. HKUST said it demonstrated a 100-pixel artificial eye in 2020 and expanded it to 367,500 pixels in this research, but the announcement does not specify mass-production processes, a commercialization timeline or operational performance in real-world systems. How the increase in pixel count translates into manufacturing yield and system-level utility will be the next point of evaluation.

Details

The HKUST team developed a three-dimensional laser-processing method to form high-density photosensitive pixels on a hemispherical substrate. The artificial retina has a layered structure with photosensitive layers sensitive to blue, green and red light, and an overall spectral response range of 300–800 nm. It produces aberration-corrected images across a field of view exceeding 160 degrees.

Sources

Translated from the Japanese original: /news/honkon-kagidai-gaku-zenshoku-gazou-to-ibento-kudougata-no-do-f7d671. The Japanese version is authoritative if they differ.