Line-of-Sight Guidance: Learning to Look Ahead in Three Dimensions

Abstract

This letter investigates line-of-sight (LOS) guidance algorithms for three-dimensional path-following. We prove that a spatial LOS guidance algorithm ensures input-to-state stability (ISS) of the closed-loop system with respect to the lateral velocity. Building on this theoretical foundation, we propose an enhanced LOS algorithm where the lookahead distance is parameterized using a neural network. This approach optimizes performance based on vehicle states and local path characteristics, which serve as inputs to the neural network, while preserving the stability guarantees. The effectiveness of our proposed method is validated through a simulation study using a high-fidelity six degree-of-freedom model of an autonomous underwater vehicle (AUV), demonstrating improved path-following performance while maintaining the stability guarantees of the original approach.