Neural Network Controller for Attitude Control of Quadrotor

Inspired from complex human brain architecture, parallel processing and learning capabilities, Artificial Neural Networks (ANN) has emerged as a new data processing paradigm. Image Association, Image Recognition, Function Approximation, Beam Forming are some of the fields that appealed many researchers to use Neural Networks. Generalization capability of Neural Networks also makes it appurtenant to Plant Control as superior control strategy than traditional Proportional-Integrator-Derivative (PID) Controllers. The research aims to design Neural Network architecture to identify whether this strategy can be employed to control a non-linear dynamic model. For this purpose well-known Vertical Take Off and Landing (VTOL) vehicle known as Quadrotor is used as a plant. Direction Cosine Matrix Inertial Measurement Unit (DCM IMU) Theory is used for attitude estimation of the flying robot. Neural Network Predictive Control and Model Reference Adaptive Control (MRAC) strategy is selected to achieve attitude control of Quadrotor. Levenberg-Marquardt algorithm is used as backpropagation technique for training of the neural network architecture. Quadrotor plant modeling, Neural Network training and an effective controller is implemented in Matlab Simulink environment and their pros and cons are reviewed. Project files are open source and available publicly on to fuel further research in area.