Rocket Thrust Vectoring Attitude Control based on Convolutional Neural Networks

Rodolfo Garcia-Rodriguez, Iván Martinez-Perez, Luis Enrique Ramos-Velasco, Mario A. Vega-Navarrete

Abstract


Launching and landing rockets on the Earth and in space have had intensive research and development in the last years. The idea of reducing costs is related regularly to the reusability of some mission stages. Though the launch has attracted attention due to including the main engines fundamental to boosting spacecraft onto an orbital or interplanetary trajectory, the landing takes relevance in space missions. While the rocket's landing has been carried out on Earth using an autonomous spaceport drone ship, it is challenging to design intelligent model-free systems that can continuously learn and compensate for slight deviations until they meet the target. This paper focuses on studying vertical rocket landing using convolutional neural networks. Assuming that the rocket is near the landing area, an attitude rocket control is proposed using a vision system to recognize it and drive -the nozzle TVC. Experimental results show the attitude control commanded by a nozzle TVC of an experimental rocket under different conditions.

Keywords


Rocket Thrust Vector Control; Convolutional Neural Networks; Attitude Rocket Control

Full Text: PDF