Robotic systems present physical limitations in the maximum torque that the motors can apply and its variation rate. Also, due to safety regulations, there are strict limitations in the maximum speed and acceleration that any link can operate. To solve these problems online the paper proposes the implementation of a multivariable constrained predictive controller. These limitations are usually very strict, hence sometimes it may not be possible to calculate an adequate sequence of the future controls due to constraint incompatibility, and the optimization problem is infeasible. To solve this difficulty an infeasibility solver can be included in the predictive controller.