Real-time Optimal Entry Guidance Using Koopman Operator Based Dynamics

Atmospheric-entry guidance poses significant challenges due to highly nonlinear dynamics, tightly coupled constraints, and strict real-time requirements. This paper presents a real-time optimal entry guidance framework that leverages a neural-based Koopman model to obtain a lifted linear representation of atmospheric-entry dynamics. The proposed method jointly learns the lifted observables, linear dynamics, and constraint mappings, enabling the inherently nonlinear entry guidance problem to be reformulated as a convex optimization that can be solved reliably and with low computational cost. Simulation studies show that the learned model accurately predicts long-horizon trajectories and constraint evolution, while the resulting convex guidance produces solutions comparable to conventional nonlinear programming methods but with dramatically reduced computation time. The proposed framework enables high-fidelity, real-time entry guidance suitable for onboard trajectory planning in challenging planetary-entry scenarios.