| Summary: | Mars represents one of the most important targets for space exploration in the next 10 to 30 years, particularly because of evidence of liquid water in the planet's past. Current environmental conditions dictate that any existing water reserves will be in the form of ice; finding and sampling these ice deposits would further the study of the planet's climate history, further the search for evidence of life, and facilitate in-situ resource utilization during future manned exploration missions. This thesis presents a suite of algorithms to help enable a robotic ice-prospecting mission to Mars. Starting from visual odometry---the estimation of a rover's motion using a stereo camera as the primary sensor---we develop the following extensions: (i) a coupled surface/subsurface modelling system that provides novel data products to scientists working remotely, (ii) an autonomous retrotraverse system that allows a rover to return to previously visited places along a route for sampling, or to return a sample to an ascent vehicle, and (iii) the extension of the appearance-based visual odometry pipeline to an actively illuminated light detection and ranging sensor that provides data similar to a stereo camera but is not reliant on consistent ambient lighting, thereby enabling appearance-based vision techniques to be used in environments that are not conducive to passive cameras, such as underground mines or permanently shadowed craters on the moon. All algorithms are evaluated on real data collected using our field robot at the University of Toronto Institute for Aerospace Studies, or at a planetary analogue site on Devon Island, in the Canadian High Arctic.
|
|---|