A decentralized algorithm utilizing the Cubature Kalman Filter (CKF) is proposed to address the multi-mobile robot localization problem in environments where GPS is unavailable, referred to as Decentralized Cooperative Localization Algorithm Based on CKF (DCL-CKF). In DCL-CKF, each mobile robot does not need to store the measurements but only maintains the latest estimate of itself. Information exchange only takes place between two robots when they obtain relative measurements of each other, saving storage costs and improving communication efficiency. Furthermore, by selecting a set of cubature points to approximately calculate the mean and covariance of the estimated state of the mobile robot, the error in linearizing the nonlinear observation function is reduced, thereby improving the localization accuracy. Monte Carlo simulations and a group of real-world experiments were conducted to verify the performance of the proposed DCL-CKF approach. The results show that the DCL-CKF can obtain consistent localization state estimation of mobile robots, and the average localization accuracy is improved by 21.68% and 32.25% compared with the decentralized Extended Kalman Filter (EKF) localization algorithm, respectively.