Virtual environments, and specifically motion-based driving simulators, raise the delicate question of multisensory cueing in order to produce the good "illusion" to the users. If virtual reality is well acknowledged for cost limitation, validity issue of these research (or training) tools is critical for knowledge transfer of new results.As a first step in the validation process of the IFSTTAR motorcycle simulator, this work relies upon an original design based on a double approach : by element and global. It consists in (i) optimal balancing of simulator’s immersive and interactive characteristics linked to roll motion so as to yield a believable and embraced leaning illusion, and (ii) evaluating more generally different simulator configurations (vehicle dynamic models) through objective (training length) and subjective measures (presence, simulator sickness).Taken as a whole, the seven studies conducted during this thesis enable the validation of the relevant range of roll motion cueing to produce a leaning sensation without leading to a critical unbalance or fall sensation. A leaning angle of the simulator beyond 11.4 degrees is not recommended to avoid any fall sensation, however this figure can be influenced by various factors (e.g., visual informations, roll axis location). These studies also allow to identify, in passive driving situations, the necessary parameters as well as the relative contribution of visual and inertial informations for the production of a believable leaning sensation. In an active driving situation, the evaluation of the degree of users control through different (virtual) vehicles dynamic models has permitted to highlight the weaknesses of IFSTTAR simulator in order to set several guidelines for further development.
