Genetic programming-based online landing guidance for a Vertical Takeoff Vertical Landing launch vehicle

Kurzfassung

This work investigates the application of Genetic Programming (GP) to the landing guidance problem of a Vertical Takeoff Vertical Landing (VTVL) rocket. Although several studies have explored the use of Machine Learning (ML) algorithms for GNC applications, most have focused on Reinforcement Learning or Neural Network-based algorithms, while little attention has been given to GP. GP offers significant potential for GNC applications, since, in contrast to other ML-based approaches, it produces white-box models that are human readable. Moreover, unlike optimization-based approaches, GP generates a guidance model that can be used online at the onboard computer frequency with a negligible computational cost. The GP algorithm is trained and validated in a 3DoF simulation environment that includes attitude kinematics, atmospheric and aerodynamic perturbations. A Monte-Carlo campaign is conducted to assess robustness and overall performance, with a LQR controller used as benchmark. The results demonstrate the viability of the proposed method, which achieves performance comparable to to that of a traditional combined scheme based on the tracking of a reference optimal solution through an LQR control architecture.