Neural-network-based position control of elastic multi-link manipulators
In this thesis a new approach of nonlinear position control is applied to structurally elastic multi-link manipulators. The general discrete dynamics of structurally elastic manipulators are first derived. Then the dynamical neural networks is introduced to achieve the position control of the manipulators. The neural-network-based controllers are trained on-line to control the manipulators. During on-line control, the coefficients of the neural-network-based controllers are also adapted based on the difference between the defined trajectory and the output of the elastic manipulators. Analysis and computer simulations are done to show the effectiveness of the design. Different conditions are used to test the neural-networkbased controllers.