University of Waterloo

Purpose

Artificial Intelligence (AI) based systems that have the ability to detect novelty is important because unexpected movement or the presence of unfamiliar objects are highly relevant for movement planning. In this case, the use of drones for infrastructure inspection is explicitly dependent on a robust novelty detection system where it is much more important to detect that something is wrong than it is to successfully classify exactly what the problem is. This Project seeks to explore and evaluate a variety of novelty detection algorithms, including both traditional machine learning algorithms and the more biologically-inspired algorithms. The Project will evaluate the computational cost of running these algorithms online, the ability of these systems to learn on-line to adjust their expectations, and the robustness of their novelty detection on infrastructure-relevant tasks. Direct applications of this system would be logistics infrastructure inspection, including railway lines, bridges, shipping containers and transmission-line inspections.