Résumé : Classical visual servoing approaches tend to constrain all degrees of freedom (DOF) of the robot during the execution of a task. In this article a new approach is proposed. The key idea is to control the robot with a very under-constrained task when it is far from the desired position, and to incrementally constrain the global task by adding further tasks as the robot moves closer to the goal. As long as they are sufficient, the remaining DOF are used to avoid undesirable configurations, such as joint limits. Closer from the goal, when not enough DOF remain available for avoidance, an execution controller selects a task to be temporary removed from the applied tasks. The released DOF can then be used for the joint limits avoidance. A complete solution to implement this general idea is proposed. Experiments that prove the validity of the approach are also provided.