This work proposes two approaches to automatic tuning of PID position controllers based on different global optimization strategies. The chosen optimization algorithms are PSO and MOPSO, i. e. the problem is handled as a single objective problem in the first implementation and as a multiobjective problem in the second one. The auto-tuning is performed without assuming any previous knowledge of the robot dynamics. The objective functions are evaluated depending on real movements of the robot. Therefore, constraints guaranteeing safe and stable robot motion are necessary, namely: a maximum joint torque constraint, a maximum position error constraint and an oscillation constraint. Because of the practical nature of the problem in hand, constraints must be observed online. This requires adaptation of the optimization algorithm for reliable observance of the constraints without affecting the convergence rate of the objective function. Finally, Experimental results of a 3-DOF robot for different trajectories and with different settings show the validity of the two approaches and demonstrate the advantages and disadvantages of every method.