Genetic Programming Guidance for the Reentry Trajectory of the ReFEx Vehicle

Abstract

In this work Genetic Programming (GP) is used to obtain an alternative guidance law for online trajectory adaptation of the Reusability Flight Experiment (ReFEx) reentry vehicle. GP is an Evolutionary Algorithm (EA) capable of producing interpretable mathematical models that satisfies user defined objectives and constraints. It can be applied in a guidance setting, through direct interaction with the environment, learning to evolve a the guidance law to steer the vehicle towards the successful satisfaction of the mission. To the best of the authors' knowledge, the application of GP to a flight-ready vehicle represents a first in this domain and can help in increasing the TRL of this technique for Guidance and Control (G&C)applications. Furthermore, it produces interpretable models, in contrast to other Machine Learning (ML) approaches, and it can be applied to nonlinear models, in contrast with traditional methods.In this study, GP is applied offline to design an feedback guidance law to guide the ReFEx vehicle towards a desired final position in the presence of uncertainties in the physical models, navigation and control signals. The performance of the GP-based guidance law is compared against the current design, which defines the trajectory correction as an optimal control problem and reduces it to a nonlinear unconstrained optimization problem, which is then solved using a successive linearization strategy. The performance comparison between the two methods utilizes high fidelity 6-DoF Monte Carlo simulations. Capabilities, advantages and disadvantages of both methods are discussed and conclusions on the viability of the investigated GP approach are drawn.