Neuro-dynamic programming applied to helicopter flight control

Successful applications of neural-control to date have been limited to simple systems, typically those possessing only a single control and a handful of states. The purpose of this paper is twofold. First it introduces to the helicopter flight control community an on-line learning control scheme, namely neural dynamic programming (NDP). Second it demonstrates that NDP performs exceedingly well as a learning controller for practical systems with higher dimensions, such as helicopters. Our discussion in the paper is focused on providing a viable alternative helicopter flight control design approach rather than providing extensive comparisons among various available controllers. The paper consists of a comprehensive treatise of NDP and extensive simulation studies of NDP designs for controlling an Apache helicopter under different flight conditions. All of our designs are tested using FLYRT, a sophisticated industry-scale non-linear validated model of the Apache helicopter.