A Probabilistic Approach to Autonomous Planetary Science with Prime and Scout Rovers

Autonomous science augments the capabilities of planetary rovers by shifting the identification and selection of science targets from the operator to the rover itself. This shift frees the rover of unnecessary idle time as it waits for commands. Additionally, as features can now be evaluated based on their scientific value, data collection can become more selective---allowing only the most valuable data to use the limited bandwidth between the rover and operator.;In addition to the development of an autonomous science system, one of the primary contributions of this thesis is an investigation into the effectiveness of pairing a larger prime rover with a smaller scout rover for improved autonomous science. In this scenario, the scout rover provides the prime rover with a science-driven feature map that can be used for path-planning, obstacle avoidance and science target selection. This scenario was compared to the traditional scenario of the prime rover solo in a simulated environment, which revealed the strengths and weaknesses of the two approaches.;The probabilistic identification and evaluation of objects using the Bayesian network and evaluation algorithm was validated experimentally in a real environment. Quantitative results measure the effectiveness of the approach while qualitative results reveal its behaviour and indicate possible areas of improvement. Together, the simulated and real tests provide suggestions for future research.;This thesis presents an autonomous science system that is comprised of three components: a Bayesian network that uses image data to identify features, an evaluation algorithm that selects the best identified feature, and a path-planning algorithm that guides the rover to the most scientifically valuable feature. As autonomous science is a relatively young field, efforts to date include partial implementations (e.g., identification or sorting without evaluation) or lack a machine learning component.