Online reentry trajectory planning for reusable rockets based on intelligent phase transition

This paper researches the online three-dimensional entry trajectory optimization for the reusable rocket. Unlike the majority of existing studies that focus on the terminal vertical landing phase, typically within the altitude range of 0 to 4 kilometers, this paper extends the trajectory to a broader range, from 4 to 80 kilometers. The wider range leads to greater nonlinearity and coupling, presenting a comprehensive multi-phase trajectory optimization framework. Unlike other studies following the typical reentry optimization scheme for glide reusable vehicles, we take account of the actual powered and unpowered descending process and propose an online multi-phase hybrid optimization strategy. The reentry process is first divided into three subphases, that is the powered descent phase, the low dynamic pressure descent phase, and the aerodynamic descent phase. The first subphase is designed for decelerating the rocket with the minimum cost of propellant, which is accomplished with model predictive control. The second subphase approximates free falling for the engine shuts down and the aerodynamic drag force is tiny. The third subphase finally satisfies strict terminal state constraints by quickly solving a nonlinear planning problem based on a novel return profile and a set of analytic solutions. Meanwhile, determining the time of phase-transition is crucial to multi-phase trajectory optimization. We solve this issue using the maximum entropy reinforcement learning method based on a developed fast reachable area prediction algorithm. On the basis of Falcon 9 (Version 1.2, Block 5), we carry out numerical simulation with the presence of random disturbances to demonstrate the method performance.