This project develops an autonomous path following capability (in the form of a sensor and algorithm kit) for aerial inspection robots used to remotely survey structures in sectors such as oil & gas, mining, energy, chemical processing, water and transport. Aerial robots have enormous potential to slash costs relative to manual inspections, which are equipment and manpower intensive and typically represent a large proportion of the recurring cost of a structure over its lifetime. Current generation robots are typically operated manually within line of sight of a remote operator; this project will develop a sensor and algorithm kit enabling such robots to automatically retrace their steps around a known structure using vision and learning, greatly speeding up repetitive surveys. A 3D visual feature map is generated and refined, and over subsequent missions a robot would use this map of the structure for autonomous visual navigation using a relocalisation approach, allowing it to reach and return from the areas to be inspected autonomously. The proposed robot combines the real-time full 3D visual mapping and relocalisation methods developed at the University of Bristol and flight control technology developed by Blue Bear.