Online Adaptive Incremental Reinforcement Learning Flight Control for a CS-25 Class Aircraft

Online reinforcement learning for fixed-wing aircraft longitudinal control

“…One of the first implementation cases for linear aircraft control simulation can be found in [18]. For nonlinear flight control, successful implementation cases have been reported for a full scale model of: missile [19,20], fixed-wing aircraft [21][22][23][24][25][26], and helicopter [27].…”

“…The longitudinal Dryden gust model used in this paper is defined by the transfer functions shown in Eq. 26…”

Reinforcement Learning based Online Adaptive Flight Control for the Cessna Citation II(PH-LAB) Aircraft

“…Incremental model techniques approximate the original nonlinear dynamical system to linear time varying system around an operating point using first order Taylor series expansion. ADP methods are combined with incremental approach to design optimal controllers suitable for nonlinear systems referred to as Incremental Approximate Dynamic Programming (iADP) controllers [17,18]. As these iADP controllers use only observed data for achieving the control iADP controllers can be classified as model free methods that has online learning capability.…”

Online Adaptive Helicopter Control Using Incremental Dual Heuristic Programming

“…These frameworks learn an incremental model in real time and therefore do not require an offline learning phase. The feasibility of this approach was demonstrated in [24], where it was shown that the IDHP framework could be used for near-optimal control of a CS-25 class fixed-wing research aircraft without prior knowledge of the system dynamics or an offline learning phase. Compared to small, fixed-wing aircraft in cruise, rotorcraft have relatively slow control responses and are unstable or marginally stable in almost all flight regimes.…”

“…Compared to small, fixed-wing aircraft in cruise, rotorcraft have relatively slow control responses and are unstable or marginally stable in almost all flight regimes. One design choice in [24] traded speed of convergence away for increased learning stability. In online adaptive control of rotorcraft, this trade-off is non-trivial, as there is the possibility of the system itself diverging before the controller has learned to control it.…”