Multivariable feedback particle filter

• Tao Yang ^[1] ; Richard S. Laugesen ^[2] ; Prashant G. Mehta ^[3] ; Sean P. Meyn ^[4]
  1. [1] GE Global Research,USA
  2. [2] University of Illinois, Urbana–Champaign, USA
  3. [3] Illinois, Urbana–Champaign, USA
  4. [4] University of Florida,USA

  Mostrar afiliaciones +
• Localización: Automatica: A journal of IFAC the International Federation of Automatic Control, ISSN 0005-1098, Vol. 71, 2016, págs. 10-23
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • This paper presents the multivariable extension of the feedback particle filter (FPF) algorithm for the nonlinear filtering problem in continuous-time. The FPF is a control-oriented approach to particle filtering. The approach does not require importance sampling or resampling and offers significant variance improvements; in particular, the algorithm can be applied to systems that are not stable. This paper describes new representations and algorithms for the FPF in the general multivariable nonlinear non-Gaussian setting. Theory surrounding the FPF is improved: Exactness of the FPF is established in the general setting, as well as well-posedness of the associated boundary value problem to obtain the filter gain. A Galerkin finite-element algorithm is proposed for approximation of the gain. Its performance is illustrated in numerical experiments