Support vector machine based model predictive adaptive control of robotic arms

For the correct calculation of the dynamics of the robotic arms and the adaptation of the controller parameters; the accuracy and precision of the model has great importance. Artificial neural networks and support vector machine algorithms are often preferred for the accurate definition of nonlinear system dynamics. Support vector machines are one of the most effective regression techniques among machine learning methods. Within the scope of this thesis, a method that will provide support vector machine based adaptive model predictive control of robotic arms is proposed. First, dynamic model estimation was performed with support vector regression using the data of a sampled manipulator. While estimating the dynamic model, the learning parameters of the trained model are optimized to prevent memorization of the training data, in other words, over-learning. The manipulator used is a four-axis lightweight robot arm. It has been observed that modeling errors and disruptive effects are minimized by using this predicted model in the adaptation mechanism. It was observed that the obtained control structure showed a successful trajectory tracking performance on different trajectories and under variable load conditions. It has been shown by the simulation studies that the proposed control structure is successful in the tracking of complex trajectories. Trajectory tracking is provided with minimum position error, especially when the time-varying load is dominant, which affects the success of the proposed controller trajectory tracking. The combined use of the model predicted by support vector regression and adaptive control mechanism is very effective against unmodeled dynamics, uncertainties, external disturbances and parameter changes in the system.