Abstract : This thesis deals with sensor-based control of robots. For a robot arm or a mobile robot, this approach consists in defining robotic tasks directly in the sensor space. This scheme is known for its good properties of accuracy and stability. The robot may still have an undesired behavior when the motion to perform is large or when constraints have to be respected. Classical schemes uses supplementary degrees of freedom to perform the main task while respecting the constraints. First we propose a new framework for sensor calibration from velocity measurements. The same formulation allows to estimate both intrinsic and extrinsic parameters of an exteroceptive sensor. Simulations and experiments illustrate this approach in the case of camera calibration. A generic framework is then exposed for multi-sensor-based control under several constraints. All constraints are introduced in the control law in the same way. The induced behavior is the respect of all constraints, while perturbing the main task as little as possible. Our formalism is applied to classical issues in robotics, that are joint limits avoidance and the visibility constraint in visual servoing. Numerous constraints are then combined into a single system.