Researchers at The Hong Kong University of Science and Technology (HKUST) have developed a high-resolution bionic eye that integrates full-color imaging, an aberration-corrected field of view exceeding 160 degrees, and event-driven motion detection. Inspired by the curved retina of biological eyes, the system offers a promising pathway toward compact vision sensors for future robots and autonomous systems, enabling machines to capture environmental information while significantly reducing the data required for motion recognition.
Led by Prof. FAN Zhiyong, Chair Professor of the Department of Electronic and Computer Engineering and the Department of Chemical and Biological Engineering, and Director of State Key Laboratory of Displays and Opto-Electronics at HKUST, the team developed a novel 3D laser fabrication approach that enables the fabrication of dense arrays of light-sensitive pixels on a hemispherical surface. The resulting bionic eye features a hemispherical tandem artificial retina with a pixel density of 1,905 pixels per inch (ppi) and 367,500 total pixels, overcoming a key resolution bottleneck for curved image sensors. The findings were published in the international journal Nature Materials in a paper titled “An aberration-corrected bionic eye for full-color vision and event-driven motion detection“.
Overcoming a key manufacturing challenge of curved artificial retinas
Vision sensors are essential components that enable machines to perceive and interact with their environment. Conventional cameras typically rely on flat image sensors, requiring multiple optical elements and external processing to correct optical aberrations. These requirements often increase system complexity and size, reduce power efficiency, and restrict the field of view.
Curved image sensors offer an alternative. By following the geometry of a biological retina, they can support compact, wide-angle imaging systems with enhanced optical performance. However, fabricating high-density pixel arrays on highly curved surfaces remains difficult, as conventional lithographic processes are primarily designed for flat substrates.
To overcome this limitation, the HKUST team developed a 3D laser fabrication approach on hemispherical substrates. This breakthrough enabled a substantial increase in pixel count, from just 100 pixels in the team’s 2020 Nature paper to 367,500 pixels in the current system, marking an important step toward the practical application of hemispherical vision sensors.
Prof. Fan said, “Our first-generation bionic eye, developed in 2020, contained only 100 sensing pixels. In this latest study, our 3D laser-assisted fabrication process enabled us to increase the number of sensing pixels to nearly 400,000 pixels, while equipping the device with an ultra-wide 160-degree field-of-view and event-driven motion detection. With continued advancements, I believe we can develop a next-generation bionic eye with over one million sensing pixels, opening new opportunities for practical applications in robotics and autonomous driving.”
Integrating full-color imaging and wide-angle vision to significantly reduce motion-sensing data
The newly developed bionic eye utilizes a hemispherical tandem artificial retina incorporating blue-, green- and red-sensitive layers. The device provides full-color imaging, with an overall spectral response spanning 300 to 800 nanometers, while maintaining aberration-corrected imaging across a field of view exceeding 160 degrees. By integrating high pixel density, full-color and wide-angle imaging into a single platform, the researchers integrated several key capabilities required for compact artificial vision systems.
Beyond image capture, the tandem design enables in-sensor, event-driven motion detection. While conventional frame-based approaches obtain motion data by processing sequences of complete images, the new system instead enables motion-related information to be detected within the sensor, eliminating the need to transmit full image sequences for evaluated tasks. Under the study’s test conditions, this approach reduced bandwidth demand by more than 99.95% compared with frame-based imaging, while achieving a motion recognition accuracy of 98.6%.
These results highlight the potential of integrating image capture and motion detection at the sensor level. Such technologies could help future robots and autonomous systems process motion-related information about moving objects while reducing visual data transmission and processing demands, paving the way for more efficient machine-vision systems.
The first author of the study is Dr. LONG Zhenghao, a former PhD student and former postdoctoral researcher in HKUST’s Department of Electronic and Computer Engineering (currently an Assistant Professor at Fudan University). Prof. FAN Zhiyong served as the sole corresponding author, and the research was conducted in collaboration with Prof. ZI Yunlong of HKUST (Guangzhou).
