Blurry light from lens imperfections is a problem everywhere from microscopes to telescopes to smartphone cameras. Using a tiny yet carefully engineered optical element and artificial intelligence, University of California San Diego engineers have built a way to spot and correct those distortions from a single image — a step that could make advanced optical systems faster, smaller and easier to use.
“We used a combination of fundamental physics, nanofabrication and machine learning to make hidden distortions easier to detect and correct,” said senior author Abdoulaye Ndao, an electrical and computer engineering faculty member in the Jacobs School of Engineering and an affiliate of the Qualcomm Institute at UC San Diego.
“Our fast, robust solution is tiny and easy to integrate into different optical systems,” he continued. “The weight is almost nothing, because the size of the sample can be one by one centimeter and half a millimeter thick.”
The work — whose co-first authors were UC San Diego Electrical and Computer Engineering Ph.D. students Sina Moayed Baharlou (also of Boston University) and Muhammad Waleed Khalid — was published in Nature Communications on May 12, 2026.
Most current methods for correcting small imperfections that can warp light and reduce image quality require several measurements, extra hardware or repeated calculations to figure out what went wrong. That can make systems bulkier, slower and harder to integrate into compact devices.
To tackle this problem, the UC San Diego team paired an optical element designed with artificial intelligence (Figure 1) with an AI-powered analysis system. Together, they give each optical distortion a unique image signature, allowing a deep neural network, a machine learning model inspired by the human brain, to read distortions from a single image of the light pattern and tell the system how to correct them. Once the researchers determined the approach could work at one wavelength, they kept expanding the device’s capabilities until it worked across multiple wavelengths, held up under noisy conditions and handled complicated beam shapes.
And the team did not just simulate the work. Using UC San Diego Qualcomm Institute’s Nano3 cleanroom facility to create the optical device and different surfaces, the researchers fabricated key optical components and tested the approach experimentally across a range of conditions.
The authors note that the paper establishes a scalable and practical foundation for real-time aberration correction for next-generation optical and photonic systems. The approach (patent pending) could improve the optical hardware used in fields including biology, astronomy and precision manufacturing.
In addition to Ndao, Baharlou and Khalid, the paper “ An end-to-end hybrid deep-learning approach for single-shot wavefront sensing and correction ” was authored by Guli Gulinihali, Jeongho Ha, Liyi Hsu and Yeshaiahu Fainman of UC San Diego (Electrical and Computer Engineering), Samantha C. Lewis of UC Berkeley, and Lei Tian and Alexander V. Sergienko of Boston University.
Nature Communications
Imaging analysis
Not applicable
An end-to-end hybrid deep-learning approach for single-shot wavefront sensing and correction
12-May-2026
The authors declare no competing interests.