Guaranteed safety and stability neural guidance for asteroid landing under collision avoidance constraints

Asteroid landing missions hold significant scientific values for deep-space exploration, ranging from scientific sampling to planetary defense mission. While existing trajectory guidance algorithms, including polynomial guidance, sliding mode guidance and convex optimization face limitations in landing safety assurance, which is the main interest of actual engineering implementation. To address the landing safety issues, we propose a criterion function control framework for safe control commands generation under uncertain asteroid dynamics and collision-avoidance constraints. The framework introduces two key innovations: Firstly, the analytical control barrier function generated online enhances control flexibility under irregular terrains with strict safety assurance. Secondly, a robustness neural control Lyapunov function trained across potential asteroid parameters is employed to ensure control stability under environment uncertainties. The landing results show 12.4% - 22.2% fuel savings compared to traditional collision avoidance methods without compromising safety. In addition, the extensive simulations highlight the effectiveness and broad applicability under various landing problems. • An online analytical criterion function generation scheme to guarantee landing safety theoretically. • A control Lyapunov network training scheme to enhance control robustness under uncertain dynamics. • A hybrid criterion control framework to increase computational efficiency while satisfying landing criterion.