Abstract : In this manuscript, we consider the problem of mobile robots navigation using multisensor- based control. Our objective is to safely perform vision-based displacements through poorly known indoor environments which may be evolutive and cluttered. The missions to be realized consist in positioning the vehicle with respect to a landmark or a person of interest. To do so, different problems have to be addressed : the motion towards the goal, the obstacle avoidance, the occlusion management, the realization of long range vision-based displacements. If the first three mentioned points are typically local issues, the last one requires to give the robot global skills. Our works have dealt with each of these problems, but, in a first step, we have only considered local issues. We have thus developed an image based visual servoing (IBVS) allowing to make the robot converge towards either a landmark or a person of interest, thus performing the nominal vision-based task. However, although this kind of control is known for its nice robustness properties, it does not allow to efficiently treat the problems of collisions and occlusions. Therefore, our next contributions have consisted in designing multi-sensor-based control strategies guaranteeing non collision when the environment is cluttered with non occluding obstacles. Then, in the sequel of these results, we have addressed the problem of the image features loss. We have designed algorithms allowing to reconstruct the visual signals when an occlusion occurs. These algorithms have then been coupled to the above mentioned control strategies, allowing to safely perform the navigation task amidst both occluding and non occluding obstacles. At this step, the mission can be realized only if the goal can be perceived from the robot initial configuration, which is not the case for a long range navigation. Our last contributions have tried to answer this problem by giving the vehicle the required global skills. To do so, we have coup led a topological map representing the environment to a supervision algorithm managing the control strategy. In this way, the robot is able to perform vision-based long range displacements in an environment cluttered with both occluding and non occluding obstacles. These works have been validated in simulation and implemented on our robots. The obtained results have shown the efficiency of the proposed approach and have opened new interesting research axes.