Physics Informed Neural Network for Optimal Rendezvous Guidance of Low-Thrust Cubesats

Orbit transfers and autonomous rendezvous are fundamental to modern space missions. With the miniaturization of electric propulsion systems, CubeSats are increasingly being considered for complex operations such as formation flying and space debris removal. However, designing optimal rendezvous trajectories for these low-thrust systems remains challenging due to their long burn durations and intricate orbital dynamics. This paper investigates the use of Physics-Informed Neural Networks (PINNs) for optimizing rendezvous trajectories of CubeSat-scale electric propulsion systems. The system is first modeled using equinoctial orbital elements and Gauss’ Variational Equations, then formulated as an optimal control task. A PINN-based approach is employed, where a neural network is trained with a physics-informed loss function to generate optimal trajectories. The Theory of Functional Connections (TFC) framework is used along with PINN to ensure strict adherence to boundary conditions. The results are compared with a numerical direct collocation method, demonstrating that PINNs compute solutions more efficiently while generating smoother trajectories that better adhere to the system’s dynamics. These findings underscore the potential of PINNs for real-time trajectory optimization, contributing to the advancement of autonomous spacecraft guidance.