On a dark road, puddles and asphalt are sometimes hard to tell apart even to the human eye. Because most existing image sensors also recognize objects mainly by differences in light intensity, they have had limits in precisely grasping surface texture, directionality, and fine structures.
Seo Jun-gi, a professor in the Department of Chemical and Biomolecular Engineering at the Korea Advanced Institute of Science and Technology (KAIST), and his team said on the 12th that they developed a next-generation sensor technology that can overcome these limits. The technology is a "self-reconfiguring polarization sensor array" that not only senses light intensity but also detects the direction in which light oscillates, and autonomously changes its operating state to find the optimal response.
Polarization refers to the property of light oscillating in a specific direction. While a typical camera mainly looks at how bright or dark light is, a polarization sensor also reads changes in the direction of oscillation that occur as light is reflected or transmitted. This information is useful for identifying the condition of a surface, material, and directionality of an object. For example, objects that reflect light, such as water, glass, and metal, can show differences in polarization information, enabling the capture of richer details that conventional sensors can easily miss.
The team used a "heterostructure" that combines two materials, tellurium and rhenium disulfide. A heterostructure is a structure that stacks materials with different properties like layers to realize new functions. The two materials respond to light differently depending on the orientation of their crystals, and the researchers leveraged this characteristic to sensitively detect the directional information of light.
In this structure, when light enters, charges move or remain at specific positions at the interface where the two materials meet. In this process, a "bipolar photoresponse" appears, where the direction of current flow changes depending on the intensity, wavelength, and direction of light. Using this, unlike conventional sensors that require external electrical signals to adjust their state, the sensor's operating mode can be altered by light itself.
This technology can be used not only for simple data collection but also for "in-sensor computing," which processes some of the sensed information on the spot. In-sensor computing processes information inside the sensor instead of separating sensors and processing units. It can reduce data movement and computational load, making it suitable for implementing low-power, high-efficiency artificial intelligence (AI) systems.
In real experiments, the team confirmed that this sensor array recognized moving objects with more than 95% accuracy. This suggests potential applications in enabling Autonomous Driving vehicles to more precisely understand road conditions and helping medical imaging equipment more accurately distinguish fine differences in tissues.
Regarding the study, Professor Seo said, "By leveraging polarization information, we presented a new foundation for AI vision technology that can secure richer visual information than before," adding, "It is expected to play an important role in implementing low-power, high-efficiency AI systems going forward."
The results were published in April in the international journal "Nature Sensors."
References
Nature Sensors (2026), DOI: https://doi.org/10.1038/s44460-026-00057-9