In this thesis, a universal trajectory planning using neural networks and evolutionary algorithm is presented and learning structures using neural- based controller is compared and modified. In the case of being only given starting point and final point, proposed universal trajectory planning provides the (sub) optimal trajectory for given cost functions through multiobjective optimization evolutionary algorithm and also determines the configurations of the robot manipulator along the trajectory by considering the robot dynamics. Neural-based controller, obtained using trajectory planning, adaptively compensates for the uncertainty and mismatch between the model dynamics and the real dynamics. Different learning structures using feedback-error learning is compared. In order to overcome drawbacks of conventional neural-based compensator using multilayer perceptrons, a new neural-based compensator uses radial basis function networks, which are a type of local neural network. The best advantage of local neural networks is rapid convergence. The effectiveness and efficiency are demonstrated through simulation studies.