Evolutionary particle filter applied to leader-labor multi-robot localization for communication failure and kidnapped situations

Abstract

For many multi-robot tasks such as surveillance, environmental monitoring, target tracking, search and rescue missions, accurate multi-robot localization is vital. We introduce a novel scheme called Leader-lobar Localization (LlL). The proposed scheme allows to formulate multi-robot cooperative localization under communication failures as a jointly self-localization and target-tracking problem. A further assumption is then made that a labor robot is kidnapped. The paper firstly examines generic particle filter (PF) and extended Kalman filter (EKF) for LlL, in which the leader robot localizes itself and use relative observations to track other team members. Additionally, as a second contribution, we introduce a novel PF called evolutionary assisted particle filter (EAPF) applied to LlL under kidnapped situation. The performance of generic PF and EKF as well as EAPF is examined using synthetic data generated according to a realistic mobile robot model. Simulation results demonstrate under kidnapped situations, the pose estimation errors for generic PF and EKF diverge, but EAPF estimates accurately the poses of robots.