Event-triggered reinforcement learning control for the quadrotor UAV with actuator saturation

This paper proposes an event-triggered reinforcement learning (RL) control strategy to stabilize the quadrotor unmanned aerial vehicle (UAV) with actuator saturation. As the quadrotor UAV equips with a complex dynamic is difficult to be model accurately, a model free reinforcement learning scheme is designed. Due to the practical limitation of actuators, the end of controller is constrained with a bounded function. In order to reduce the calculation consumption for the onboard computer, an event-triggered mechanism is developed, which only update the controller when the triggered condition is satisfied. The proposed controller is implemented with two neural networks which are called critic and actor. Some advanced RL technologies are utilized for speeding up the train process, e.g. off-policy training, experience replay, etc. The stability of closed-loop system is proved by the Lyapunov analysis. The simulation results including a stability task and a tracking task verify the theoretical analysis, in which we find the updating frequency of controller is decreased greatly.