Reinforcement Learning-based Control Allocation for the Innovative Control Effectors Aircraft
Pieter S. de Vries, Erik-Jan Van Kampen · AIAA Scitech 2019 Forum · 2019
Established Control Allocation (CA) methods rely on knowledge of the control effectiveness for distributing control effector utilization for control of (overactuated) systems. The Innovative Control Effectors (ICE) aircraft model is highly overactuated with its 13 control effectors, CA is a preferred method to distribute control effector utilization. In this paper it is envisioned to use Reinforcement Learning (RL) for distributing control effector utilization, which requires no knowledge of the control effectiveness. RL allows to pursue more abstract and timescale separated objectives. The ICE aircraft’s altitude, considering only longitudinal motion, is controlled by distributing the control effector utilization using RL, from an initial offset, while pursuing secondary objectives such as decreasing effector utilization and thrust.