Better Robots See like a Spider

Dan Drollette

How do you increase resolution and decrease fixed pattern noise in a robot's vision while using fewer photoreceptors? By borrowing an idea from the tropical jumping spider, say researchers Oliver D. Landolt and Ania Mitros. Mitros is a member of Christof Koch's laboratory team at California Institute of Technology's Center for Neuromorphic Systems Engineering, and Landolt is a former member.

The lab draws upon animal models to solve engineering problems. When Koch heard that a jumping spider can see as well as a human -- even though it has only 800 photoreceptors in each of its retinas, in contrast to the 137 million photoreceptors in our eyes -- he saw it as a model to emulate.

The spider's secret? Its vibrating retinas, which constantly sweep from side to side to scan visual information. This allows its photoreceptors to pinpoint where changes in light intensity occur. In contrast, fixed photoreceptors can overlook these changes if they occur in the spaces between photoreceptors.

The electronic image sensor system subsequently built by Landolt and his colleagues similarly vibrates. The CMOS chip they designed contains photoreceptors, or pixels, arranged in a 32 x 32 array attached to a metal frame. A lens is fixed at one end. The whole unit is mounted on springs, which causes it to shake each time the robot moves. Each shake shifts the chip's pixels to cover more area.

Because the lens-to-chip distance remains the same, the image stays in focus. No lateral motion can blur the image because each pixel independently generates and processes a series of very fast digital pulses -- in a sense, each pixel makes a high-speed movie consisting of many separate freeze-frames.

A signal processor then takes each pixel's output and combines them, factoring in their changing positions over time, to build an image.

Landolt said that besides increased resolution, their vibrating chips offer other advantages. Because preprocessing is done on the chip, smaller and less power-hungry microprocessors are required. This means that lighter robots consuming less electricity can be built.

He said NASA's Jet Propulsion Laboratory in Pasadena is considering using the chips in the visual navigation systems of the next Mars rovers. And they could also work in autonomous underwater vehicles.