Robotic rock classification and meteorite search

Liam Pedersen
Carnegie Mellon University

Friday, July 23, 1999
12:00 Noon—Pacific Forum

Pedersen nomad.jpg (11246 bytes)

Modern unmanned planetary rovers suffer the serious shortcoming of being unable to return more than a fraction of the total scientific data that they are capable of acquiring. This is due not only to bandwidth and storage limitations, but also because the robot can only intermittently download data and receive new commands. Given these limitations, a robot able to process the scientific data, obtained so as to recognize objects of interest (rocks, meteorites, fossils etc.) and survey an area without direct supervision from mission control, would have a clear advantage. More samples could be obtained, larger areas surveyed, and a smaller workload placed on mission control (itself very significant for a mission lasting several months). The terrestrial applications for operations in remote environments, such as Antarctica or underwater, are clear.

A key ability for robotic geological exploration is the ability to autonomously recognize rocks. This talk will describe a Bayesian solution to the problems of recognizing different rock types from a mobile robotic platform, and making generalizations about their geographical distribution. This goes well beyond simple pattern classification, as rock classes are not well defined. Many different sensors are required, yet their use must be minimized to conserve vehicle resources. The ultimate dream is to imbue a robot with the abilities of a competent field geologist, particularly that of being able to recognize scientifically significant rocks and minerals, and to notice the exceptions to the rule, such as meteorites, fossils, or a green rock on Mars.

The system described has been partially implemented on Carnegie Mellon University’s Nomad robot vehicle (which recently returned from field tests in Antarctica) where it is to be used to search for meteorites. Details of the Robotic Antarctic Meteorite Search (RAMS) program are online at http://www.frc.ri.cmu.edu/projects/meteorobot/.

Next: Shiner perch divergence and the phylogeny of the Embiotocidae

 Last updated: December 19, 2000