AI to help Martian exploration

By Paul Rincon
BBC News Online science staff

The cyborg astrobiologist consists of a wearable computer and camera

Scientists are developing a wearable, "intelligent" computer system to help humans or robots explore Mars on future missions to the Red Planet.

The "cyborg astrobiologist" consists of a person equipped with a compact computer, camera, finger mouse, head-mounted display and keyboard.

It is designed to help explorers select interesting geological features for investigation on the Martian surface.

The system is currently designed for a human, but could be used with a robot.

Field trials

Scientists are carrying out field trials with the system in the countryside of Central Spain.

They have also tested it on a poster of the Martian surface taken by the Mars Pathfinder probe in 1997. The system successfully picked out edges of a hill on the horizon and rocks with more saturated colours.

The 667 MHz wearable computer runs on a "power-saving" processor and is worn on a special belt strapped to the wearer's back.

A head-mounted visual display, camcorder, keyboard and finger mouse completes the system. The finger mouse allows the user to navigate in Microsoft Windows.

Rocks (left) and a segmentation map of the same scene (right)

Principal investigator Patrick McGuire, of the Centro de Astrobiologia in Madrid, Spain, told BBC News Online:

"If you're trying to develop a robot and its computer vision software at the same time, then you always have to worry about the robot breaking down.

"I wanted to develop a system that I could test more quickly without having to worry about the robot."

'Interest map'

The intelligent program produces separate images showing the hue, saturation and intensity of a given scene.

These "derived" images are then divided into different "segmented regions" based on features in the image. Each of these segments is then assigned a number based on how large or small it is.

The smaller regions tend to contain the most uncommon features and are therefore considered most interesting.

For example, a medium-sized red rock on an otherwise barren surface of brown dirt might be the uncommon region of an image.

The visual display has dimensions of 640x480 pixels

Three "uncommon maps" are produced for each of the derived images, which are then combined to produce an "interest map".

"The user can define the initial values to give it a bias towards edges or a bias towards polysaturated pixels," Dr McGuire explains.

This allows the user to tailor the system to their own needs or to the quirks of a particular landscape.

Dr McGuire is currently working to produce flow charts based on geological analysis by co-investigator Jens Ormo.

These can then be "translated" into algorithms capable of being used by the cyborg astrobiologist, giving it a foundation in geological method.