Deep neural networks based Gravitational Field for Asteroid Landing Control

The irregular shape of the asteroid makes it challenging for precise gravitational field modeling, further impacting control precision and making onboard computational time intolerable for the control law. To enhance the efficiency and accuracy of gravitational field calculations and control for asteroid landing, deep neural networks (DNN) are employed to approximate the asteroid's gravitational field. To this end, the DNN is trained by utilizing samples generated from the polyhedral model. Then, the asteroid gravitational field DNN is applied to multiple sliding surfaces control for asteroid landing. Finally, numerical simulation results of landings on 433 Eros are presented, demonstrating that the DNN yields highly accurate approximations and significantly improves the efficiency of control law resolution.