You-Jeong LEE, Dain YOON, Se-Hwan AN, Chang-Hun LEE, Hyoung-Sik CHOI, Seungchan SHIN
DOI Number: N/A
Conference number: HiSST-2025-253
This paper presents a sampling-based discrete model predictive control (KL-MPC) framework for space-plane equipped with on/off thrusters. In contrast to conventional Model Predictive Path Integral (MPPI) control, which perturbs continuous inputs with Gaussian noise, the proposed approach represents actuator commands using Bernoulli distributions and updates their probabilities through cost-weighted rollouts. This discrete formulation inherently accommodates binary firing logic without heuristic thresholding, while actuator dynamics are captured through a first-order lag model. The cost function is evaluated in moment-tracking coordinates to simultaneously account for reference tracking accuracy,
fuel efficiency, and switching reduction. Simulation studies on a reentry spaceplane demonstrate that the proposed method mitigates overshoot and inefficiency associated with pseudo-inverse thresholding, while achieving performance comparable to an optimization-based MPC–MILP controller at submillisecond runtimes. The results demonstrate that KL-MPC achieves high-performance attitude control while remaining computationally efficient and explicitly respecting discrete actuator constraints.