Independent space navigation is a reality

We at ProcessIT are passionate about innovation in automation and digitalization. Here is an interesting example of research on IT security in space – and how the solution can be found in automated AI.
 
The space sector is developing at an extraordinary pace. More and more industrialization challenges will arise that need to be solved to advance the shared mission of developing the future space infrastructure and its operations. Adam Abdin and his team in Paris are creating an AI-driven solution for autonomous collision avoidance maneuvers in space.
 
– The goal of our research is to successfully introduce higher levels of automation for key Space Traffic Management (STM) processes, particularly collision avoidance with space debris or another spacecraft. This is to ensure the level of reliability needed for navigating many spacecraft in an increasingly complex environment, Adam Abdin Associate Professor at CentraleSupélec.
 
Traditionally, experts have been responsible for planning and executing spacecraft collision avoidance maneuvers (CAMs), a process that typically takes hours to days of preparation. However, due to the growing demands of Space Traffic Management, there is a need for intelligent onboard autonomous systems to handle spacecraft maneuvering tasks rapidly, robustly, and efficiently.
 
Although Space Situational Awareness (SSA) and Collision Avoidance (CA) technologies have improved, they still heavily rely on human decision-making. This strategy may not be sustainable as the space environment becomes more complex. The research aims to develop robust Artificial Intelligence frameworks that enable autonomous decision-making of spacecraft navigation and maneuvering equivalent to that of human experts, if not better.
 
– By using Reinforcement Learning it’s possible to tackle uncertainties in spacecraft decision-making, modeling it as a Partially Observable Markov Decision Process (POMDP). Another solution we created is an algorithm based on Deep Recurrent Q-Network (DRQN) capable of solving the proposed POMDP with continuous and infinite state space and discretized action space, Adam Abdin states.