Entry trajectory generation based on neural network

Abstract

A methodology for onboard generation of entry trajectory subject to all common inequality and equality constraints is developed, which makes use of the neural network as a major approach to design a complete and feasible entry trajectory instantaneously. Conventional constrained nonlinear trajectory optimization problems and control parameters generation online can be transformed into the neural network off-line training problem, given the entry initial conditions, values of constraint parameters, and final conditions. Differing with the general neural network, this approach is trained by the principles of optimal theory. The inputs of the neural network are the time-variant state variables, the outputs are the near optimal control parameters. Numerical simulations with a reusable launch vehicle model for various entry conditions are presented to demonstrate the capability and effectiveness of the approach.