CAUSAL LEARNING IN UNMANNED/AUTONOMOUS VEHICLE DYNAMICS

Abstract

With the ascent of unmanned aerial vehicles (UAVs) in commercial and research sectors, addressing their susceptibilities to wind disturbances becomes paramount. Current methodologies, though effective, often hinge on specialized sensors, thereby adding to the UAV's weight and compromising its functionality. This thesis explores a recently proposed approach called "causal curiosity", using machine learning (ML) to identify varying wind conditions solely from a UAV's position trajectory, circumventing the need for dedicated wind speed sensors. Through the application of time series classification combined with the intrinsic "causal curiosity" reward system, the research delves into discerning three distinct wind environments: constant wind, shear wind, and turbulence. Ultimately, autonomous UAVs can employ this paper's findings to design optimal trajectories in challenging weather conditions.