Human-Mars Entry, Powered Descent and Landing Guidance

Learning-Enabled EDL Guidance: We combine machine learning with optimal control theory to generate onboard guidance commands with low computational cost. Rather than learning complete state and control trajectories, our approach identifies a compact set of parameters that capture key features of fuel-optimal solutions and trains neural networks to predict these parameters from vehicle states. The predicted features are then used to reconstruct continuous-time guidance commands, reducing the dimensionality of the learning problem. This framework extends from powered descent to the complete entry, powered-descent, and landing mission, with multi-stage updates enabling rapid guidance regeneration in response to disturbances. Simulation studies demonstrate the potential to achieve precise, fuel-efficient landing while substantially reducing online computation.

Selected Publication:

1. S. You*, C. Wan*, R. Dai, and J. Rea, "Onboard Fuel-Optimal Guidance for Human-Mars Entry, Powered-Descent, and Landing Mission based on Feature Learning", Acta Astronautica, Vol 195, pp. 129-144, 2022. DOI: 10.1016/j.actaastro.2022.02.007

2. S. You*, R. Dai, and J. Rea, "Theoretical Analysis of Fuel-Optimal Powered Descent Problem with State Constraints", Journal of Guidance, Control, and Dynamics, Vol. 45, No. 12, 2022, pp. 2350-2359, DOI: 10.2514/1.G006815

End-to-End EDL Guidance: We optimized atmospheric entry, powered descent, and landing as an integrated mission rather than treating each phase independently. A unified multiphase formulation jointly determines entry maneuvers, powered-descent initiation conditions, and thrust commands while accounting for phase-specific dynamics and operational constraints. To solve this nonconvex problem efficiently, we develop a customized alternating direction method of multipliers (ADMM) that exploits a quadratic reformulation of the trajectory optimization problem. Numerical studies demonstrate fuel savings compared with separate-phase guidance, highlighting the importance of coordinating atmospheric deceleration and propulsive braking to improve overall mission performance.

Selected Publication:

C. Wan*, G. Jing*, R. Dai, and J. Rea,"Fuel-Optimal Guidance for End-to-End Human-Mars Entry, Powered-Descent, and Landing Mission", IEEE Transactions on Aerospace and Electronic Systems, Vol.58, no. 4, pp. 2837-2854, 2022. DOI: 10.1109/TAES.2022.3141325

Six-Degree-of-Freedom Entry Guidance: We developed entry guidance methods that account for both the translational and rotational motion of human-scale Mars vehicles, moving beyond point-mass trajectory models. Using dual quaternions, we represent position and attitude in a unified formulation that simplifies the coupled dynamics. The resulting optimization jointly plans vehicle motion and control moments to minimize speed at the end of atmospheric entry while satisfying dynamical, operational, and mission constraints. A hybrid ADMM algorithm efficiently solves the associated nonconvex quadratic optimization problem, with numerical studies validating the modeling approach and computational advantages. This work connects trajectory planning with attitude dynamics to support more physically realistic entry guidance.

Selected Publication:

C. Pei*, C. Wan*, R. Dai, and J. Rea, "A Hybrid ADMM for Six-Degree-of-Freedom Entry Trajectory Optimization based on Dual Quaternions", IEEE Transactions on Aerospace and Electronic Systems, Vol. 59, No. 3, 2023, pp. 3280-3295, DOI: 10.1109/TAES.2022.3223333

Powered Descent Guidance with Landing Site Selectiion: We develop powered descent guidance methods that jointly select a safe landing site and generate a fuel-efficient trajectory, instead of assuming a fixed landing site. Our mixed-integer optimal control formulations combine discrete choices among candidate landing sites with continuous trajectory and thrust optimization while accounting for terrain hazards and vehicle constraints. A multiphase iterative algorithm addresses the resulting nonconvex optimization problem through a sequence of convex subproblems. To enable rapid onboard guidance, we further develop mixed-input learning methods that combine terrain-map images with vehicle position and velocity to predict key parameters of the optimal solution. These parameters support simultaneous landing-site determination and guidance generation without repeatedly solving the full optimization problem onboard.

Selected Publication:

1. C. Pei*, S. You*, D. Yu, and R. Dai, "A Multi-Phase Iterative Algorithm for Mixed-Integer Optimal Control", Journal of Guidance, Control, and Dynamics, Vol. 48, No. 4, 2025, pp. 757-770. DOI: doi.org/10.2514/1.G008165

2. C. Pei*, S. You*, and R. Dai, "Mixed-Input Learning for Multi-point Landing Guidance with Hazard Avoidance Part I: Offline Mission Planning based on Multi-Stage Optimization", AIAA Science and Technology Forum and Exposition, National Harbor, MD, 2023.

2. S. You*, C. Pei*, V. Kenny*, and R. Dai, "Mixed-Input Learning for Multi-point Landing Guidance with Hazard Avoidance Part II: Learning-base Guidance Algorithm", AIAA Science and Technology Forum and Exposition, National Harbor, MD, 2023.