Radiation Aware Navigation and Planning for Mobile Robot Exploration of Unknown Hazardous Nuclear Environments

This thesis presents a unified framework for radiation-aware robotic exploration and navigationfordeploymentonresource-constrainedroboticplatformsoperatinginhazardous nuclear environments. Conventional navigation frameworks remain largely geometry-centric and do not explicitly account for radiation exposure, increasing the risk of hardware degra- dation, reduced mission endurance, and mission failure. Central to the framework is a sparse occupancy-map abstraction method that con- verts dense occupancy grids into lightweight node–edge networks whilst preserving naviga- tional connectivity. The abstraction reduces computational complexity by up to two orders of magnitude, enabling scalable real-time planning on embedded robotic hardware. Building upon this representation, a radiation-aware planning framework is devel- oped that integrates hazard-weighted node scoring and modified A* heuristics to minimise cumulative radiation exposure during navigation. Experimental evaluation demonstrates an average 85.6% reduction in cumulative radiation dose with only a 17.2% increase in traversal distance relative to conventional distance-optimised planning. To enable operation in previously unknown environments, the framework is extended through online Gaussian Process Regression (GPR) for real-time radiation field esti- mation and adaptive exploration. Results demonstrate reductions in cumulative radiation exposure of more than 75% compared to radiation-unaware exploration whilst maintaining complete environmental coverage. To bridge the gap between simulation and deployment, this thesis additionally introduces a novel ultrasonic beacon-based radiation emulation system capable of reproducing inverse-square radiation field behaviour in a safe and repeatable manner. The platform enables realistic validation of radiation-aware robotic systems without radioactive materials and demonstrates two orders of magnitude greater stability than comparable radio-frequency surrogate approaches. Finally,thecompleteframeworkisdeployedandvalidatedonaUnitreeGo2quadruped robot using a ROS 2-based architecture integrating SLAM, radiation emulation, probabilis- tic radiation estimation, and hazard-aware planning. Together,thesecontributionsestablishascalableandexperimentallyvalidatedframework for radiation-aware robotic autonomy, advancing the state-of-the-art in safe autonomous operationfornucleardecommissioning, hazardous-environmentinspection, disasterresponse, and future planetary exploration missions.