Analysis of selection methods for cost-reference particle filtering with applications to maneuvering target tracking and dynamic optimization

Cost-reference particle filtering (CRPF) is a recently proposed sequential Monte Carlo (SMC) methodology aimed at estimating the state of a discrete-time dynamic random system. The estimation task is carried out through the dynamic optimization of a user-defined cost function which is not necessarily tied to the statistics of the signals in the system. In this paper, we first revisit the basics of the CRPF methodology, introducing a generalization of the original algorithm that enables the derivation of some common particle filters within the novel framework, as well as a new and simple convergence analysis. Then, we propose and analyze a particle selection algorithm for CRPF that is suitable for implementation with parallel computing devices and, therefore, circumvents the main drawback of the conventional resampling techniques for particle filters. We illustrate the application of the methodology with two examples. The first one is an instance of one class of problems typically addressed using SMC algorithms, namely the tracking of a maneuvering target using a sensor network. The second example is the application of CRPF to solve a dynamic optimization problem.