This paper considers the problem of cooperative localization (CL) usinginter-robot measurements for a group of networked robots with limited on-boardresources. We propose a novel recursive algorithm in which each robot localizesitself in a global coordinate frame by local dead reckoning, andopportunistically corrects its pose estimate whenever it receives a relativemeasurement update message from a server. The computation and storage cost perrobot in terms of the size of the team is of order O(1), and the robots areonly required to transmit information when they are involved in a relativemeasurement. The server also only needs to compute and transmit update messageswhen it receives an inter-robot measurement. We show that under perfectcommunication, our algorithm is an alternative but exact implementation of ajoint CL for the entire team via Extended Kalman Filter (EKF). The perfectcommunication however is not a hard requirement. In fact, we show that ouralgorithm is intrinsically robust with respect to communication failures, withformal guarantees that the updated estimates of the robots receiving the updatemessage are of minimum variance in a first-order approximate sense at thatgiven timestep. We demonstrate the performance of the algorithm in simulationand experiments.
展开▼
