Researchers at the Beckman Institute's AI Group and Robot Vision Lab have developed a new camera system that uses both audio and visual cues to point itself toward "interesting" activity. The camera is controlled by a neural network that simulates the superior colliculus of the human brain. A wide angle camera and two microphones are used to localize activity and a second camera with a tighter view is then directed towards the desired area. The system contains a database of sounds to help it determine what types of sound are interesting and, over time, it learns to associate sounds with the visual targets that make them.