The persistent localization of moving targets is one of the most important applications of mobile robot systems. However, the sensors on each individual robot are often not sufficiently accurate for this task, presenting a severe limitation on the application of robots to many situations. To overcome this problem, systems of multiple robots have been established. These systems improve the localization accuracy of mobile targets by fusing together multiple sensing data. Numerous studies have demonstrated that the relative motion (or specifically, the relative pose) between the robot and the target, as well as that among different robots, strongly influences the final localization accuracy of the data fusion algorithm. In other words, the regulation of motion in a multiple-robot system will lead to better localization results. Thus, in this paper, a new active persistent localization (APL) scheme is proposed. The underlying concept of this scheme is based on set-membership descriptions of uncertainties. The basic problem is formulated in the framework of an enhanced set-membership filter, and a new data fusion algorithm with improved accuracy and convergence properties is designed. A motion-planning algorithm is then incorporated under the so-called optimal localization condition to complete the APL scheme. Simulations using multiple 3-D mobile robot systems and experiments on a multiple rotor-flying-robots test bed show that the proposed algorithm successfully enhances the accuracy of the persistent localization of the moving target.