Optimal dynamic Control Allocation with guaranteed constraints and online Reinforcement Learning

Abstract This paper introduces a new formulation of dynamic control allocation as a dynamic optimization problem. This optimal control formulation allows us to develop allocation with guaranteed actuator constraints, and to learn the optimal control allocation online using measured data and without knowing the system dynamics. The general solution to this problem is provided in the form of an H ∞ controller. Current results for static control allocation are shown to be a special case of the general dynamic optimization solution. Reinforcement learning is used to find the optimal solution for the constrained actuators problem. The methods proposed in the paper are tested on a F-16 flight simulation.