A new multi-robots cooperative hunting strategy based on Cross-EKF localization is proposed for enhancing convergence rate, robustness and precision. In this strategy, the posterior estimate covariance for target location estimated by multi-robots is crossly calculated, and a minimum covariance is obtained. The maximum distance from edge points to mean center is used as radius to construct a convergence circle. The convergence to dynamic point is expanded to circle surface. The experimental results show that the circle is quickly and smoothly converged and the target is accurately hunted. The method possesses high practical value.