Autonomous planetary rovers require obstacle detection capabilities to navigate hazardous terrain without Earth-based intervention, yet currently deployed methods are limited. This research develops and validates a complete deep learning system for autonomous rock detection on a resource-constrained planetary rover.
We present a lightweight MobileNetV2-based U-Net architecture with dual attention mechanisms, optimized for edge deployment with only 0.31 million parameters. A new dataset MarsTanYard was created via a semi-automated dataset creation pipeline, enabling efficient annotation of Mars-analogue terrain imagery. Our deep learning network was trained on this dataset and integrated with a ROS2 navigation stack through a modular architecture that transforms segmentation masks into 2D occupancy grids that can be used for rover path planning. Our network achieves a 77% intersection-over-union accuracy for rock segmentation, and physical validation on a testing rover in a Mars analogue environment demonstrates a 94% detection rate for large rocks at close-range. An inference time of 4.49ms was achieved on the target rover hardware using model optimization techniques. The system maintains reliable operation across varying lighting conditions with less than 15% performance degradation.
Results show theoretical collision probability of 7.8 × 10^-7 per rock encounter, enabling months of autonomous operation for typical planetary missions. This work provides an end-to-end validation of the deep learning obstacle detection system, establishing a foundation for enhanced rover autonomy in future Mars exploration missions.

The growing intensity and frequency of fires and volcanic activity have heightened the demand for real-time monitoring of thermal events. This Thesis presents the first comprehensive end-to-end processing pipeline for real-time segmentation of thermal hotspots in raw multispectral imagery. The pipeline combines an onboard Deep Learning (DL) design with the use of raw imagery to achieve real-time performance. The processing pipeline was thoroughly tested on CubeSat edge computing hardware, demonstrating its feasibility for CubeSats with limited computing resources. The DL architecture used was specifically designed in this project to address onboard constraints such as model size, complexity and inference time. The satellite imagery used in this work is from the Sentinel-2 mission, and a key contribution of this project is the creation of SegTHRawS, the first dataset for thermal hotspot segmentation in raw multispectral imagery.

As the Moon reemerges as a renewed fronteer in space exploration, the Lunar Zebro project proposes to deploy a swarm of miniature rovers for efficient lunar surface exploration. One of their goals is to leverage recent advancements in deep learning and AI-accelerating hardware, in conjunction with Commercial Off-The-Shelf technologies and the NewSpace movement, to enhance the autonomous capabilities of these nano-rovers. This research focuses on integrating AI-accelerating hardware within the stringent Size, Weight, and Power (SWaP) constraints of these lunar rovers. It evaluates the suitability of various hardware configurations. A Convolutional Neural Network for hazard detection was trained and tested on different devices and scenarios. Finally, the operational cycle of the rover was simulated and the constrained resources were tracked for the different design options.