Convergence, stability and robustness of learning control schemes for robot manipulators