Robotics applied to rehabilitation requires specific manipulators: Powered Orthoses. They are orthopedic devices equipped with motors and captors that enable locomotor assistance. These powered orthoses must be capable of reproducing physiological articular trajectories and taking over or simulating the segmentary charges of a movement, mainly walking. One needs to obtain rather high dynamic performances with the help of small activators enabling mechanical integration bearable for its user. This doctoral thesis deals with the conception of such POs as well as it proposes original parallel kinematics that are compared with serial kinematics, i.e. ordinary exoskeletons. We have chosen to limit our study to motor re-education of lower limbs associated to neurological disorders: paratetraplegia, hemiplegia, cerebral palsy, etc. This project was initiated by the Fondation Suisse pour les Cyberthèses (Swiss Foundation for Cyberthoses) in 1999, in collaboration with the Laboratoire de Systèmes Robotiques (Laboratory of Robotic Systems) of the Ecole Polytechnique Fédérale de Lausanne (EPFL). It aims at developing systems of motor re-education and walking assistance, associating powered orthosis with trans-cutaneous and closed-loop electrical muscle stimulation: the MotionMaker™ and the WalkTrainer™. Its goal is to create active muscular participation that respects body dynamics of movements and to quicken re-education, whenever possible. In cases of total paralysis, its purpose is to activate lower limbs in order to reduce side-effects and complications resulting from immobilization. The FSC also makes it its ambition to conceive powered orthoses for autonomous walking with functional electric stimulation in its research programme: the WalkMaker™. To begin with, this doctoral thesis defines the biomechanical bases relative to lower limbs and pelvis. Anthropometrical, kinematic and dynamic data of body segments, as well as space-time parameters of walking have been specified. These data have been used in the theoretical models of the conception of the Powered Orthoses, and applied to the numerical simulations carried out in this study. Then we have presented the state of research on orthoses. The number of projects and the diversity of technologies offer a good illustration of the challenge posed when conceiving Powered Orthoses. So far there are no autonomous Powered Orthoses for re-education of ground walking subsequent to neurological trauma. If it is possible to find treadmills on the market, they are however deficient for medical purposes because of the subject's passivity and lack of pelvis mobility. We have then conceived two Powered Orthoses for a stationary training device: firstly, a serial orthosis of the exoskeleton type has been conceived with three articulations: hip, knee and ankle. Its activators consist in connectingrod and crank systems. Secondly, we have devised a device of leg manipulation with a parallel structure (in the shape of the Greek letter lambda λ). We have modelled these two powered orthoses and carried out a numerical simulation of two movements in order to compare their performances: leg press and cycling. The λ parallel powered orthosis gives better results. Before these findings, we built the prototype of the serial powered orthosis for clinical tests for feasibility of closed-loop controlled electric stimulation. The thesis then offers the design of an unprecedented powered orthosis to be integrated into a walker. We have analyzed and modelled a parallel structure with orthogonal connections and another one with λ connections. The comparative results of the numerical simulations of normal speed walking show that the orthogonal powered orthosis is optimal for an autonomous walker. We have built a prototype, and walking tests with healthy subjects prove the feasibility of such a concept. Finally we carried out two studies for a leg-powered orthosis, compatible with a pelvic PO and the walker. An exoskeleton is compared with a parallel structure. For these two systems, the numerical simulations and models give all the kinematic and dynamic features of the activators. Following these results, we chose the parallel PO so as to design an experimental prototype. It is currently being built as we are writing these lines. To complete the procedures, a chapter deals with the integration of the POs into a walker, with the motorization of the mobile frame and with a system of active body relieving. We also present a system of optical measurements of pelvic movements for diagnosis or for biomechanical studies. The last chapter offer guidelines for development in powered orthosis.