Design of a Nonlinear Controller on Quadrotor Drone Using Combined Method of Gradient Particle Swarm Optimization

In the paper a new method of optimal control in presented which is composed of policy gradient reinforcement learning and particle swarm optimization. This method has a lot of applications in the real world. The combined method is implemented on a quadrotor drone to control attitude and position of the drone. Inspired from reinforcement methods, the gradient of the policy is computed for a proportional-integral-derivative controller and used in particle swarm optimization to be used in optimization process in addition to the other factors. To study the performance of Optimal proportional-integral-derivative controller on attitude control of the system, a quadrotor is fixed to the design a test stand. The system consists of an accelerometer and a gyroscope sensors and a microcontroller which is used to design fuzzy proportional-integral-derivative attitude controller for the quadrotor. Considering that the experimental data has lots of errors and noises, Kalman filter is used to reduce the noises. Finally using Kalman filter leads to better estimation of the quadrotor angles and the optimized proportional-integral-derivative controller performs the desired motions successfully. The presented method is implemented and tested on the quadrotor test bench and compared with some old methods. To check the robustness of the proportional-integral-derivative controller to the external disturbances, random disturbances are applied to the quadrotor. The controller stabilized the quadrotor rapidly even with disturbance is applied.