abstract: Solving multi-objective problems is achieved using interacting multi agent systems. In this talk, present different agent dynamics which aim to address all minimization problems simultaneously. This is achieved by assigning to each agent a particular sub-problem and by evolving this label over time. This additional repulsive dynamics in the space of labels allows to improve the approximation of the Pareto front, leading to improved solutions compared with the original multi-objective problem. A second scale, the mean-field approximation of the dynamics, shows that convergence properties are preserved. To conclude, we present numerical experiments for bi- and tri-objective problems.