Researchers have developed a neuromorphic vision device that mimics one of the human visual system’s most remarkable capabilities: the ability to adapt seamlessly to dramatically different lighting conditions. The technology, described in Nature Communications, could help improve machine vision in applications ranging from autonomous vehicles to robotic systems.
Human vision can function across an enormous range of illumination levels, allowing us to move from a dark cinema into bright sunlight or navigate roads illuminated by oncoming headlights. This adaptability relies on complex biological mechanisms, including the bleaching and regeneration of photopigments in rod and cone photoreceptors. Replicating this capability in artificial vision systems has proven challenging, with most existing technologies relying on computationally intensive algorithms or complex circuitry to compensate for changing light conditions.
The new device takes a different approach. Developed by researchers from China, the UK, and the US, the system is based on a TiO₂/PEDOT photomemristor – a light-sensitive electronic component that combines sensing and memory functions. Rather than relying on external processing, the device physically adapts its own photosensitivity in response to ambient light.
The mechanism hinges on water. Under bright illumination, a photothermal effect causes water molecules to desorb from the PEDOT layer, reducing the concentration of hydronium ions and decreasing conductivity. As a result, photosensitivity is suppressed. In dim conditions, water is reabsorbed, restoring conductivity and increasing sensitivity. This dynamic adsorption–desorption cycle allows the device to self-regulate its response to light, in a manner analogous to biological visual adaptation.
According to the study authors, the approach produces an unusually strong adaptive response. Under intense illumination, the device’s steady-state signal can fall below its baseline dark current – a level of suppression not typically achieved in previous adaptive vision devices. This enables the system to attenuate overwhelming bright-light signals while preserving information from dimmer objects in the same scene.
That capability becomes particularly important in mixed-light environments, where conventional systems often struggle. The researchers use the example of a driver attempting to detect pedestrians and traffic signals while facing the glare of oncoming headlights. Existing adaptive sensors generally perform well only under relatively uniform illumination, whereas the new photomemristor can simultaneously suppress bright background light and enhance weaker visual information.
To test the concept, the team built a 4 × 4 photomemristor array and coupled it with an artificial neural network. In a simulated mixed-light environment containing dim, moderate, and bright regions, the system successfully identified image patterns after adaptation had occurred. Recognition accuracy increased dramatically as the adaptive response developed, reaching 93.7 percent in one demonstration and approximately 91.3 percent overall under mixed-light conditions. Notably, this performance was achieved without the complex image-processing algorithms typically required for such tasks.
While the technology remains far from clinical ophthalmology, the work highlights how biological principles of visual adaptation continue to inspire advances in artificial vision. By borrowing from the eye’s ability to dynamically regulate sensitivity, the researchers have demonstrated a hardware-based approach that could reduce computational burden while improving performance in real-world visual environments.
