Gaussian message-based cooperative localization on factor graph in wireless sensor networks

Abstract

Location information has become a critical requirement for many applications in wireless sensor networks. Conventional localization requires dense anchors with known positions or high transmit power in sparse networks to reach successful localization, which is not suitable for low-cost and lowpower sensors. Cooperative localization is a promising solution for wireless sensors' localization, in which the agents needing to be located cooperate with neighboring nodes by exchanging messages and perform measurements with them. In this paper, a distributed cooperative localization algorithm on factor graph is proposed to locate the sensors. Resorting to the linearization method to tackle the nonlinearity in range measurement, Gaussian parametric messages are obtained with closed forms using the sum-product algorithm on factor graph, which leads to low computational complexity and low communication overhead. Numerical simulations are performed to evaluate the proposed algorithm, which shows superior performance to the particlebased SPAWN estimator.

A Robot Web for Distributed Many-Device Localisation

“…Then Caceres et al [7] extended [44] to a network composed of GNSS nodes. These distributed positioning methods were later generalized by adding nonlinear measurement models and utilizing Gaussian message passing [33], [34]; and in 2017, Wan et al [43] proposed a distributed multi-robot SLAM algorithm, using belief propagation. In their method, a mixture of Gaussian and non-parametric models was used to handle nonlinear models.…”

Distributed simultaneous localization and mapping for mobile robot networks via hybrid dynamic belief propagation

“…In Wu et al, 27 distributed Bayesian cooperative localization was solved via Gaussian message passing on factor graph, but nonlinearity between the positions and range measurements made the optimization problem complex and multivariate; therefore, the linearization on the Euclidean norm in the range measurement is employed, which ensured that closed-form expression can be obtained to represent the messages on factor. To further improve the closed-form Gaussian representation solutions to nonlinear observation model, the Taylor expansion was used to approximate the nonlinear terms in message updating, while a broadcast message scheme approximating the exact message passing was utilized to reduce the traffic burden in Li et al 28 But what cannot be ignored was that above studies 27,28 did not give an even generalized motion model, namely, the former offered a simple linear state-space model and the later did not mention it; and what was more, they required anchors whether or not accurate although the placement of anchor nodes was normally seen, for example Wymeersch et al 29 SPAWN, a cooperative localization method based on the sum-product algorithm and factor graph, was proposed in Wymeersch et al 29 and applied to ultra-wide bandwidth wireless networks; however, it distinguished between agents with priori unknown states and anchors with known states at all times, both of which may be mobile, and became extremely difficult in directly computing messages. On the whole, some message passing algorithms over factor graphs can be responsive to distributed cooperative localization problems; going a step further, we could consider to design an inference algorithm based on factor graph model to cope with the SLAM problem, which is a relatively less practice to the best of our knowledge.…”