A comparative study of cooperative localization techniques for sensor networks

This paper focuses on the cooperative localization problem in sensor networks where the objective is to estimate the sensor positions. Communications among the sensors can reduce the need for global observations (e.g. GPS data), and render the estimation distributed. However, these problems usually involve nonlinear models and non-Gaussian distributions. Two techniques are studied and compared to address nonlinearity and non-Gaussianity: the feedback particle filter (FPF) [16] and the nonparametric belief propagation (NBP) [11]. FPF introduces a novel feedback and innovation structure, and the computations can be approximately localized. On the other hand, NBP reformulates localization as an inference problem on spatio-temporal graphical models and implements a distributed sample-based message-passing scheme. Comparisons between FPF and NBP are provided regarding their structure, computational cost and accuracy, and are supported by numerical simulations.