How to Support the Machine Learning Take-Off: Challenges and Hints for Achieving Intelligent UAVs

Abstract

Unmanned Aerial Vehicles (UAVs) are getting momentum. A growing number of industries and scientific institutions are focusing on these devices. UAVs can be used for a really wide spectrum of civilian and military applications. Usually these devices run on batteries, thus it is fundamental to efficiently exploit their hardware to reduce their energy footprint. A key aspect in facing the “energy issue” is the exploitation of properly designed solutions in order to target the energy- and hardware-constraints characterising these devices. However, there are not universal approaches easing the implementation of ad-hoc solutions for UAVs; it just depends on the class of problems to be faced. As matter of fact, targeting machine-learning solutions to UAVs could foster the development of a wide range of interesting application. This contribution is aimed at sketching the challenges deriving from the porting of machine-learning solutions, and the associated requirements, to highly distributed, constrained, inter-connected devices, highlighting the issues that could hinder their exploitation for UAVs.