On Decentralized Self-localization and Tracking Under Measurement Origin Uncertainty

Abstract

We propose an algorithm for simultaneous Cooperative Self-localization (CS) of a network of mobile agents and multi-target tracking (MTT) under complete data association uncertainty. Specifically, the associations between measurements and objects, i.e., agents and targets, are unknown. Existing CS-MTT algorithms do not assume origin uncertainty for both interagent and agent-target measurements. Due to the joint density being intractable, a message passing scheme is employed to approximately infer the marginals of agent and target states, where the number of targets is unknown and time-varying. Based on average consensus, we propose a distributed Gaussian implementation of the proposed method, which only requires communication between one-hop neighbors. Numerical experiments show the improved performance of the proposed CS-MTT algorithm as compared to the conventional approach of separate localization followed by tracking.