Thesis Oral

  • Liam Pedersen

Robotic Rock Classification and Autonomous Exploration

The capability to autonomously conduct a scientific investigation isessential to the next generation of robotic vehicles for the explorationof the solar system. Bandwidth and storage limitations mean thatcurrent vehicles, such as the Mars Pathfinder and Sojourner duo, returnonly a fraction of the scientific data they are capable of acquiring,nor can they respond to observations in real time. Future missions, suchas a robotic search for life in the oceans under the ice cap of theJovian moon Europa will have extremely limited communications withEarth, yet must operate in an unknown and dynamic environment. Therobot will need to autonomously recognize scientifically interestingoccurrences and conduct follow up observations if it deems themnecessary.A first step in the development of an autonomous science capability fora robotic explorer is a system allowing it to autonomously recognize arestricted class of scientifically interesting objects (such as rocks,meteorites or fossils) and survey an area, learning what objects arepresent and permitting the identification of anomalies.This thesis is about the application of Bayesian statistics to theproblem of autonomous geological exploration with a robotic vehicle. Itconcentrates on the sub-problem of classifying rock types whileaddressing the issues associated with operating onboard a mobile robot,and argues that the Bayesian statistical paradigm, using a Bayes networkgenerative statistical model of the geological environment, isparticularly suited to this task. This paradigm is extended in anatural way to solve the more general robotic problems of autonomouslyprofiling an area and allocating scarce sensor resources. Majorconsiderations are the need to use of multiple sensors and the abilityof a robotic vehicle to acquire data from different locations. Needlesssensor use must be curtailed if possible, such as when an object issufficiently well identified given sensor data acquired so far. Furthermore, by investigating rocks in many locations, the robot has theopportunity to profile the environment. Different rock samples arestatistically dependent on each other. These dependencies can beexploited to substantially improve classification accuracy.The classification system has been implemented onboard the Nomad robotdeveloped at Carnegie Mellon University, and applied to the task ofrecognizing meteorites amongst terrestrial rocks in Antarctica. InJanuary 2000 A.D., Nomad was deployed to Antarctica where it made thefirst autonomous robotic identification of a meteorite.Further Details:Committee Members: Martial Hebert, Chair William
For More Information, Please Contact: 
Catherine Copetas, copetas@cs.cmu.edu