Distributed object pose estimation over strongly connected networks

Abstract

We propose distributed strategies to estimate the pose, which is defined in the special Euclidean group SE(3), of a static object observed by a multi-agent system whose interaction graph is strongly connected. The individual measurement observed by each agent is assumed to be Gaussian (on SE(3)). The optimal pose of the object is computed by the agents in the network based on maximum likelihood estimation (MLE) and by using only local information exchanges between the agents. To cooperatively compute the optimal pose in a distributed manner, we first propose a distributed estimation strategy over strongly connected networks having a row-stochastic weight matrix. The distributed estimation method ensures a linear convergence of the estimated states to the optimal one. Then, the pose estimation errors of the agents are linearized by the Baker-Campbell-Hausdorff (BCH) formula. By using the linearized pose error model and the distributed linear estimation method, we finally propose distributed algorithms to estimate the optimal pose of the object. The proposed algorithms consider not only the object poses observed with regard to a global reference frame but also the object poses measured in the local reference frames of the agents. In the absence of the global reference frame, a distributed localization algorithm using noise-free relative pose information is proposed under the strongly connected network. The proposed algorithm ensures that the estimated poses linearly converge to the agent poses with regard to a common reference frame.(c) 2023 Elsevier B.V. All rights reserved.