Deep Neural Network Based Self-Tuning PID Control for Quadrotor Attitude

For decades, the Proportional Integral Derivative (PID) controllers have been successfully used in academia and industry for several types of plants. This is due to its versatility and sufficient efficiency in linear or linear plants and in such circumstances in non-linear plants. A variety of PID controller gains tuning methods have been suggested in the Quadrotor Unmanned Aerial Vehicle (UAV) literature over the last decades, several of them off-line. However,in those situations the plant is exposed to persistent parametric shifts or external disturbances, online gain tuning is a suitable alternative. This is the case for the UAV quadrotor, where the parameters linked to weight are being applied to the mass adjustment according to the system to which it is attached. In practice, some time is spent on tuning the PID gains in the UAV attitude. Once the best range of gains has been made, the UAV is able to start off smoothly. However, as the UAV changes are subject to winds, their output deteriorates as the fixed set of gains is no longer true under the current conditions. In order to solve this problem, an online PID gains tuning algorithm should be introduced. This paper proposes an auto-tune PID controller based on Deep Neural Networks (DNN). The DNN plays the function of automatically calculating the required range of PID gains to achieve device stability. The DNN changes the online benefits of the controller to obtain a smaller location monitoring malfunction. Simulation results are provided taking into account the under-implemented 6 DOF (degrees of freedom) Quadrotor UAV. Real-time tests on Quadrotor test-bench are conducted to demonstrate the feasibility of the proposed scheme