Parallel robots have good performances in terms of ability to support loads, rigidity and rapidity. However, their control and modelling are complex due to the kinematic chains closure. The use of computer vision in the feedback simpli¯es the kinematic models by introducing additional sensors into the mechanism and thus yields easier control. Several control laws are proposed in this thesis. First, a 3D pose visual servoing is proposed, where the end-e®ector pose is indirectly measured and used for regulation. It is shown to be more suitable to this task since it relieves the control from the di±cult forward kinematic problem. Second, we wanted to fuse kinematics and projective geometry into a projective kinematic model for control. Doing so, inspired by the geometry of lines and the image projection of cylindrical legs, we proposed a structure devoted to control which uni¯es several families of parallel robots. Taking this approach, we have proposed a generic projective kinematic model for visual servoing of parallel robot using legs observation. The main advantage of this method is that we do not require the use of a visual pattern to calibrate the relative pose of the latter with respect to the end-e®ector. We con¯rmed that, measuring by vision the leg edges, one can easily measure a projective kinematic model for control without proprioceptive sensors, which might simplify the robot design.
