This small sensor may enhance the ability of self-driving cars and robots to see in low light conditions.

      Researchers at Penn State have created a light-adaptive photomemristor inspired by the human eye that achieves more than 95% visual accuracy in changing light conditions.

      Penn State researchers have introduced a light-responsive sensor component that could enhance the reliability of cameras in autonomous vehicles and robots under varying lighting scenarios. This research, published on Monday in Nature Communications, draws directly from the way the human eye adapts to different levels of brightness.

      Using biology as a model

      Existing camera systems in self-driving cars are designed for stable lighting, leading to reduced accuracy when lighting conditions change quickly, such as transitioning from a dark road to bright oncoming headlights. The team from Penn State, co-led by engineering professor Larry Cheng, sought inspiration from the rod and cone cells in the eye. In the eye, rod cells contain pigments that are bleached in bright light and gradually regenerate in darkness, allowing for continuous adjustment of sensitivity.

      Replication of eye dynamics

      The researchers emulated this capability in a new kind of photomemristor, a compact sensor that captures light and converts it into electrical signals. Their design incorporates two materials: a conductive gel-like polymer and titanium oxide. When light strikes the titanium oxide, it produces a current that leads the polymer to either absorb or release water based on the light intensity, thereby self-regulating sensitivity in real-time.

      Achieving 95% accuracy in changing lighting

      To evaluate their design, the team constructed a 4×4 array of these sensors and integrated it with a neural network, forming a fundamental machine vision system. They tested it using a variation of the standard eye chart, asking it to identify an LED letter "F" against varying brightness backgrounds. After seven training cycles, the system achieved over 95% accuracy in mixed lighting conditions.

      Each sensor is only half a millimeter in width, and Cheng mentions that individual units can be connected into larger arrays to detect broader visual patterns without altering the size of each component.

      The team sees potential uses for this technology beyond autonomous vehicles, including in factory robotics and eventually assistive devices for individuals with visual impairments. A provisional patent has been filed, and next steps involve developing the sensors into a multimodal system capable of processing both visual and tactile data simultaneously.

      This photomemristor adds to an expanding array of sensor innovations aimed at enhancing the reliability of autonomous vehicles, including a compact radar system developed at Rice University earlier this year.