From robotics to information processing capability: the potential of AI for the national space ...
Artificial intelligence is pervading and profoundly changing every sector of the technology industry, including aerospace. The applications in this field are as manifold as the growth opportunities it offers the national space economy.
The development of artificial intelligence is crucial for Sistema Paese, to consolidate Italy's leading position in the space economy and to enhance the national academic community, which is expanding and has a strong international reputation.
In the current economic scenario, competition with other players in the field of foundational and technological challenges,' explains Barbara Caputo, lecturer at the Politecnico di Torino and director of the university's Artificial Intelligence Hub (AI-H@PoliTo), 'is minimal. It is therefore appropriate to intercept and seize the opportunity offered by AI. Investment in Artificial Intelligence for aerospace,' she continues, 'from foundational research to applied research and the development of prototypes and products, must be consistent, coordinated and continuous, in order to grow and protect this supply chain, which is, to all intents and purposes, a critical infrastructure for the country.
Artificial intelligence finds multiple spaces in the aerospace world. First of all in automated systems such as space robotics and unmanned aerial vehicles. The most advanced machine learning techniques are used, for example, in the use of satellite images for the early detection of forest fires, the development of computer vision algorithms for surveillance systems and for the autonomous navigation of drones, and the design of systems for robust control in the domain of robotics in a space context.
Recent applications are aimed at optimising production processes such as the Slingshot Aerospace system that helps aerospace companies manage risks and threats, and detect and map debris using predictive analytics, geospatial signal processing and computer vision. Neurala instead uses computer vision and deep learning technologies to classify images and recognise objects to autonomously avoid obstacles. SparkCognition, on the other hand, exploits machine learning to warn of aircraft and asset failures before they occur, maximising fleet availability and minimising unscheduled maintenance. Then, using natural language processing to reduce resolution time, it automatically classifies fault codes and recommends the best corrective actions. And through reinforcement learning it provides a simulated environment in which a user can test the fleet's control algorithms and tactics.
But it is above all in the collection and analysis of data that artificial intelligence demonstrates its full added value.
Today, satellites collect a huge amount of data that is sent back to Earth where it is stored and analysed, using massive cloud computing resources. Thanks to the application of artificial intelligence, on-board analysis of the collected data could be carried out, reducing the amount of data transmitted to the ground (and the costs associated with this) and at the same time allowing satellites to react in near-real time to what is observed.
An example of this application of artificial intelligence is the In-Orbit Space Lab, a multifunctional platform developed by Asi that will take small satellites into space that can generate and process information directly in orbit. In this lab, artificial intelligence,' explains Tiziana Scopa, an engineer at the Italian Space Agency, 'makes the advantages concrete in terms of optimisation of on-board resources, processing times, and adaptation for system reconfigurability. The approach is to provide the right information, at the right time, in the right place. The really important thing about this information is that it will be generated directly on board, in space, optimising resources and reducing the cost of delivering services to users. One feature of the laboratory will be this: having real-time information that does not require processing on the ground, but is directly expendable by the user at the right time. Artificial intelligence algorithms,' explains the engineer, 'trigger a data processing chain, which does not perform the canonical steps that are done on the ground, but only extracts the useful part, with mechanisms similar to those of neural networks, which therefore simulate human reasoning. It is thus possible to have the information (e.g. a fire alert) in a few dozen bytes communicated over the network, instead of gigabytes of raw data. This is a flexible and revolutionary approach compared to the traditional spatial model.
It 
In