Spacecraft autonomy challenges

The satellite could collide with space debris and lose attitude control and, in the worst case, key systems. A “smart” system could autonomously change its orbit to avoid any obstacles and simultaneously evaluate the variation of other parameters that can affect the mission (for example, high temperature variations due to sun exposure). The use of artificial intelligence in this case could help the satellite manage its position autonomously by establishing the best possible trajectory according to the variables involved.

Machine learning can help the system to memorize a series of new information and independently evaluate possible scenarios based on the variables that characterize the event. All variables are made dependent on each other through a nonlinear function, that is consolidated only when the calculation error tends to zero. The logic of machine learning and neural networks identifies problems on the basis of specific algorithms, which have been built with historic data, or real-time, and therefore capable of finding an optimal solution by reducing uncertainties. Such a system can perform or aid decision making in real time.

The manoeuvres of attitude correction are considered to maintain the satellite position with respect to an inertial reference (Sun or other fixed stars). After a perturbation, the trajectory of the satellite is not a perfect conic or does not follow an Keplerian orbit. The satellite must be regularly corrected. After having formulated a generic architecture of the control system, the compensation systems or the control systems are designed to calculate by a frequency analysis (for example the Bode diagrams) the satellite response. The use of the step response, according to the attitude parameters and the steady state error, can meet the specifications of a closed-loop project.

In this scenario, the integration of artificial intelligence into a consolidated system becomes the main function of automation understood as a technology that, through logical methods, manages new challenges, reducing the need for human intervention or for simulated responses. Artificial intelligence will be able to simulate the best response through the knowledge of numerous variables that are not necessarily dependent on each other.

Not everything can be automated, but we manage to have some events controlled by artificial intelligence. The possible decisions made in real time and the continuous monitoring of the satellites are to be considered linked to the recent development and integration of cognitive technologies in the architecture of satellite communication systems.

This technology is based on the acquisition of data at a cognitive level; through a series of sensors data is acquired and then processed according to a dedicated algorithm. Applying artificial intelligence and machine learning, satellites autonomously manage attitude control subsystems without interruption, making real-time decisions without waiting for instructions from the ground. Machine learning algorithms can identify obstacles in your path more quickly. Distant orbits can be identified, leaving the system to choose to switch from one orbit to another, to reach cosmic objects far away from Earth.

Artificial intelligence, cognitive automation and machine learning can increase the way we communicate via satellite and the way we design space technology. To analyse the data made available during the mission (internal and external) it is important to record the information in a large database which, by communicating with the main algorithm, allows autonomous decisions to be made during the mission. In this way we try to shift the decision-making process from ground to space, relegating to the ground segment the management of fault, or events related to the mission objectives. Human operators can therefore focus on more critical and more complex decisions that, for now, are out of reach for artificial intelligence based systems.