Date of Award

Spring 2024

Access Type

Dissertation - ERAU Login Required

Degree Name

Doctor of Philosophy in Aerospace Engineering


Aerospace Engineering

Committee Chair

Troy Henderson

First Committee Member

Riccardo Bevilacqua

Second Committee Member

Richard Prazenica

Third Committee Member

Sirani Perera

College Dean

James Gregory


The origin of machine learning has significantly impacted various scientific domains, offering novel solutions to longstanding challenges, particularly perception-related tasks. In space exploration, perception is a fundamental sensing capability that provides a variety of information conveyed by light and the photons that create it. This can be from identifying patterns, finding where the spacecraft is, determining chemistry composition, and classifying hazards. Perception is vital for hazard-relative navigation, mapping, and planetary landing tasks. These tasks are critical for the success of missions that require complete autonomy when exploring outer space; even on Earth, communication methods need to be faster to maintain an operator in the loop.

Nevertheless, they present complex challenges due to extraterrestrial environments' dynamic and unpredictable nature, such as illumination conditions and geography. This research is motivated by the limitations of traditional navigation and mapping techniques using passive sensors, which often need help adapting to the uncertainties and hazards in unfamiliar terrains. This thesis aims to develop and validate a comprehensive set of machine learning tools to enhance the accuracy, reliability, and efficiency of navigation, mapping, and landing processes on planetary bodies.

The methodology employed in this research encompasses designing, developing, and implementing several machine learning algorithms tailored for spatial data analysis and decision-making under uncertainty. These tools leverage advanced deep-learning techniques to interpret complex terrain data, predict hazardous zones, and help optimize landing trajectories. A significant portion of the research involved creating simulated environments to train and test the machine learning models, ensuring their robustness and reliability in real-world conditions.

The results of this study demonstrate a significant improvement in hazard detection, terrain mapping accuracy, and landing precision compared to existing methods. The machine learning tools developed as part of this thesis have proven capable of dynamically adapting to new information, enabling more effective navigation and safer landings on uncharted planetary surfaces.

This thesis's contributions extend beyond the development of machine learning tools. It provides a novel framework for integrating machine learning into space exploration missions and offers insights into the potential of machine learning to transform other aspects of space exploration, such as resource identification and astronaut assistance.

This research opens several avenues for future work, including refining algorithms to handle increasingly complex scenarios, integrating these tools into actual space missions, and exploring their applicability in other domains. This work illustrates different applications for hazard detection and avoidance using machine learning for various domains that use vision as a primary capability, such as surface rovers, robotic arms, and landers. These applications use machine learning for different purposes to enhance the data products that can be extracted from the raw vision data.

Finally, this work also addresses the implementation of some of these pipelines in embedded hardware, proving that these tools are close to a mature state that is almost compatible with the current embedded hardware and will be ready for the new space hardware developments.