Analysis of design solutions for autonomous drones
Bogdan Nedelcu, Adina Magda Florea · 2023
Drones as a particular case of robots were recently regulated by the EU as a guideline for all member states, raising several barriers in the use of drones in industrial applications for the future. Because of safety risks and privacy concerns, from the legal point of view, drones have now a limited applicability outside the visual line of sight of the human pilot therefore a limited economical applicability to some areas accessible by the pilot and outside urban areas. In this paper, we identify the key paths autonomous drone technology should progress towards, to establish a strong basis for a future certification of operating at human tolerable safety risks, and we explore the existing options to simulate in 3D environments the behavior of drones and use simulator rendered data to capture sensor data and images for constructing a set of training and validation data. Moreover, we report on the abilities of simulators to integrate software and hardware in the loop autopilots of drones, explore ways of using custom network architectures on drones, and test the ability of low power computing chips to run custom design neural network architectures. We explore the existing options to simulate in 3D environments the behavior of drones and use simulator rendered data to capture sensor data and images for constructing a set of training and validation data. We also test the abilities of simulators to integrate software and hardware in the loop autopilots of drones and possibility to integrate well known programming languages. To further explore the possibility of using custom network architectures on drones, we test the ability of low power computing chips to run custom design neural network architectures, and we identify some of the challenges ahead.