This thesis addresses the issue of maintaining information coherence and its robustness in a multiagent system, that collectively gathers information from distributed sources and where some sources may be defective (deliberately or not). In this context, the collective information gathered by the system is a progressive (possibly non-linear) aggregation of information collected individually by each agent. Therefore, each agent has direct information, collected by itself, and indirect information, obtained through communication with other agents. System coherence is defined by the compatibility of collected information about the explored environment and its actual information. System robustness is defined by the capability to maintain information coherence, despite the existence and increase of faulty agents within the system. To ensure the system coherence, this thesis proposes a trust model named TrustSets, allowing agents themselves to reason about collected information to ensure its consistency by using the calculation of the information reliability. Each agent maintains a trust network and can recognize direct (collected from the environment) and indirect (collected by exchanging information with other agents) information, not only in its stored data, but also in the data transmitted by agents it encounters. Then, the agents develop their own local and global communication strategies to ensure the system robustness against the effects of dissonance agents. To ensure the system robustness, we construct a multi-agent system which brings out dynamically strategies of movement and communication automatically adapted to the perturbation. For this purpose, we propose a self-organizational approach, based on a systemic view in which we consider a structural coupling between two levels: direct information gathering and communication. This mechanism acts as a guide for communicating and for limiting the propagation of dissonant information in the system. Consequently, it reduces the impact of dissonant information on the process of gathering information collectively. Various experiments were conducted as the part of a collaborative mapping application to show the interest of our approach.
