Researchers Simulate Fly’s Vision

Barbara Grant

Insect biologists studying the visual capabilities of flies may find their efforts aided by an innovation developed at the California Institute of Technology. Christof Koch, professor of computation and neural systems, and graduate student Reid Harrison have created a robotic model of the fly's vision system that interacts with the surrounding environment in real time, providing researchers with an alternative to complex computer simulations of fly vision.

"Approximately half of the fly's brain is dedicated to visual processing," Harrison said. The fly's visually guided behaviors include stabilizing its flight path, evading predators and chasing fast-moving airborne targets such as potential mates. It accomplishes these tasks with a visual system that has approximately five orders of magnitude fewer photoreceptors than are found in humans.

The researchers simulated the fly's visual system with a chip integrated with photodetectors and analog circuitry. The chip measures 5 x 1.5 x 1 cm and has been installed on a variety of small autonomous platforms, including one made from Lego building blocks. Each detector's response to a visual stimulus feeds into a delay-and-correlate circuit, which compares its intensity with the delayed intensity response of its neighbors. A high correlation between two adjacent detector outputs indicates target motion in one direction; a low correlation indicates motion in the opposite direction. Additional circuitry sums up the correlated outputs and sends a signal to the robot's motors, causing the device to move toward or away from the target.

The chip consumes only tens of microwatts, Harrison noted, because of its analog electronics. He believes that autonomous navigation systems in tiny air vehicles might utilize this technology, enabling flight control capability on vehicles so small that they can fit into the palm of a hand.

For the time being, Harrison is content to work with flies. He is studying the interaction between the vision system and its gyroscopes, and can envision the development of integrated smart sensors with focal-plane processing to simulate fly vision. "I'm intrigued that flies can do so much with low-resolution eyes," he said.

The researchers' work was described in the Sept. 3 issue of Science.