In this paper we study the problem of the assignment of road paths to vehicles. If we assume available the real-time road network information, then (self-concerned) vehicles select paths in a way related to user optimization which results in Wardrop equilibrium. The latter, even though fair for the vehicles of the same Origin-Destination (O-D) pair, in general can be arbitrarily more costly than the system optimum. System optimization, on the other hand, can produce unfair assignments both for the vehicles of the same as of different O-D pairs. To surmount the performance issue of the user-in respect to the system-optimization while considering the fairness issues, we propose a MAS-based distributed optimization model for path assignment to vehicles from the same and different O-D pairs at two levels. On the upper level, the proposed model optimizes the overall O-D pairs' Nash Welfare with the fairness related constraints while on the lower level, for every O-D pair separately, paths are assigned to individual vehicles through the auction algorithm. We test the solution approach through simulation, compare it with the conventional user-and systemoptimization, and thus demonstrate that it results in fair and globally efficient path-vehicle assignments.