Since swarm intelli g ence allows self-or g anization into an unfamiliar environment and adapting behaviors throu g h sim p le individuals' interactions, we p ro p ose to realize a swarm multirobot organization with a fuzzy control. We introduce in this p aper a fuzzy system for avoidin g the collaboration sta g nation and to im p rove the counter-ant al g orithm (CAA). The robots' collaborative behavior is based on a hybrid approach combinin g the CAA and a fuzzy system learned by MAGAD-BFS (Multi-a g ent Genetic Al g orithm for the Design of Beta Fuzzy System). A series of simulations enables us to discuss and validate both the effectiveness of the hybrid a pp roach to the problem of environment exploration (i.e., for the p ur pose of cleanin g an area) as well as the usefulness of MAGAD-BFS for learning the fuzzy knowled g e base while tunin g it and reducin g its number of rules.