Our sensory inputs, as well as our movements, are uncertain. Nevertheless, our central nervous systems appears to be as accurate as possible: these uncertainties are handled in an optimal fashion. For example, redundant sensory signals are weighted according to their accuracy, and motor uncertainties are taken in account when movements are made. The characteristics of the combination of redundant sensory signals are well known for static tasks. However, they are less known in dynamic conditions. The experimental part of this thesis allowed to confirm the use of statistically optimal sensorimotor processes during movements. We showed that visual information can be integrated during sequences of saccades, the oculomotor system being able to use visual information to update its internal estimate of eye position. A complementary study on the sources of variability for saccadic eye movements suggests a similar role for extra-ocular proprioception. In a third original experiment, we showed that tactile input is optimally taken in account for the on-line control of arm movements during which fingertips are in contact with a textured surface. Last, we built several neuronal networks models simulating optimal movement planning. These networks were based on current knowledge about probabilistic representations in the nervous system
