A Framework For Autonomous Spacecraft De-orbiting Taking Into Account Uncertainties

With the advent of the New Space era and mega-constellations, Low Earth Orbit (LEO) is becoming densely populated. This translates into an increasing risk of collisions in orbit, which, in turn, could threaten other space assets. This cascading effect, which peaks with the Kessler syndrome, raises significant concern in the space community. Therefore, the need for reliable collision avoidance and disposal at the end of life is more critical than ever. In LEO, the removal of a satellite is often performed by a deorbiting process, allowing it to burn up in the Earth’s atmosphere. However, simulating the trajectory at low altitudes is challenging. The complex dynamics includes Earth atmospheric drag, solar radiation pressure, as well as gravitational perturbations from the Sun and the Moon. The implementation of these effects can be tedious for actors that want to design reentry strategies, collision avoidance algorithms, and autonomous decision-making solutions. This highlights the need for a comprehensive and easy-to-use framework. Many existing software and libraries can be considered to model the satellite motion and provide implementations of the most common perturbing accelerations. Among the available solutions, the Open Source Basilisk framework stands out by its structure that is suitable to describe multiple systems in interaction and is especially suitable for hardware-in-the-loop implementations. Relying on the Basilisk simulation engine, we propose here a framework dedicated for both autonomous decision process and de-orbiting trajectory simulation. On one hand, representation of a realistic environment is required, taking into account state-of-the-art models and inherent uncertainties. In particular, we need to be able to model uncertainties in the physical models used, whether atmospheric models (unknown solar activity), irregular Earth gravity field, spacecraft sensors and actuators, or the structure of the spacecraft itself. On the other hand, the framework copes with a common decision-making API standard. For instance, relying on the Reinforcement Learning (RL) paradigm, the spacecraft can learn from the simulator a policy to ensure a reentry strategy taking into account uncertainties of the models. Simulation is scalable to ease RL computation, while hardware in the loop eases testing and validation for the on-board implemented decision policy. The final goal is to allow selection of an optimal plan or policy for deorbitation, while enabling validation possibility for an on-board implementation of the decision-making algorithm.