Neural-embedded learning control for fully-actuated flying platform of aerial manipulation system

Abstract

To address the problem of efficient task for the fully-actuated aerial manipulation system, the flight platform control scheme based on neural networks embedding is proposed. We embed the neural networks controllers (NNC) into some base controllers, which are Lyapunov stable for their system. Due to the learning and optimizing capability of NNC, the derivative controllers allow to improve control performance by updating the parameters of NNC under Lyapunov stability condition. The backstepping technology based on the mixture basis functions that approximates unknown system dynamics and the improved disturbance observer is designed for the base controller of platform attitude. And the linear active disturbance rejection controller (LADRC) is utilized as the base controller of platform position. Considering safety and cost, we made some virtual experiments in CoppeliaSim a software which offers high accuracy physical engine. The results of virtual experiments prove that compared with some state-of-the-art technologies the methods proposed are advanced in tracking errors performance due to NNC.