关键词: 重复使用火箭, 动力故障, 返回段轨迹优化, 解析规划, 强化学习

Abstract: The return process of reusable launch vehicles undergoes multiple phases, including free fall, high-altitude powered de-celeration, unpowered deceleration, and terminal powered deceleration. The corresponding trajectory planning problem is subject to stringent terminal and procedural constraints, and faces the challenge of complex energy management throughout the powered and unpowered flight phases. Specifically, power failure can significantly degrade the rocket's range capability and maneuvering envelope, leading to substantial deviations of the aerodynamic and thermal environ-ment from nominal conditions, and ultimately compromising landing safety. Extending the unpowered (aerodynamic) de-celeration phase serves as a primary approach to compensate for insufficient powered deceleration; additionally, it offers the benefits of fuel conservation and reduced engine wear under nominal power conditions. Against this backdrop, focus-ing on the high-altitude powered deceleration and unpowered deceleration phases, this paper proposes a segmented online trajectory planning strategy for the return phase under power failure. The strategy comprises three key compo-nents: a fuel-optimal powered descent phase based on convex optimization, a low-dynamic-pressure deceleration phase designed for smooth powered-unpowered phase transition, and an unpowered deceleration phase with analytic planning that automatically satisfies strict terminal handover constraints. Meanwhile, a reinforcement learning agent is trained. Ultimately, an online decision-making method for shutdown states under power failure is developed, encompassing key parameters such as shutdown altitude, velocity, and flight path angle. Finally, simulations are conducted using the overall parameters and typical return states of the Falcon 9 rocket. The results demonstrate that the proposed method can com-plete three-dimensional trajectory reconstruction for the entire return phase within 0.8 seconds, even when thrust drops by more than 30% and initial state and aerodynamic deviations are taken into account.

Key words: Reusable launch vehicle, Power failure, Return phase trajectory optimization, Analytic planning, Reinforce-ment learning