The knowledge of realistic dynamic models to robotic actuators would be of great aid in the conception of control laws for robot manipulators, mainly in the cases of the great precision robotic or still for manipulators with flexible links. In this paper we present a training strategy and propose a structure of Neural Network (NN) to learn the friction torque of a geared motor drive joint robotic actuator. To train the NN it was used a friction model proposed in the literature (published in 1995). It was considered the motor torque and the rotor angular velocity as input in the NN, while the friction torque was the only output, which was used in the proposition of a non-linear friction compensation mechanism. This compensation mechanism was used in parallel with a proportional and derivative control of a SCARA robot. The results attested the efficiency of the NN friction estimate and compensation with the proposed mechanism.
