Distributed Particle Filtering over Sensor Networks for Autonomous Navigation of UAVs

Abstract

State estimation and control over sensor networks is a problem met in several applications such as surveillance and condition monitoring of large-scale systems, multi-robot systems and cooperating UAVs. In sensor networks the simplest kind of architecture is centralized. Distributed sensors send measurement data to a central processing unit which provides the state estimate for the monitored system. Such an approach has several weaknesses: (i) it lacks fault tolerance: if the central processing unit is subject to a fault then state estimation becomes impossible, (ii) communication overhead often prohibits proper functioning in case of a large number of distributed measurement units. On the other hand decentralized architectures are based on the communication between neighboring measurement units. This assures scalability for the network since the number of messages received or sent by each measurement unit is independent of the total number of measurement units in the system. It has been shown that scalable decentralized state estimation can be achieved for linear Gaussian models, when the measurements are linear functions of the state and the associated process and measurement noise models follow a Gaussian distribution (Nettleton et al. 2003). A solution to decentralized sensor fusion over sensor networks with the use of distributed Kalman Filtering has been proposed in (Olfati-Saber 2006), (Watanabe & Tzafestas 1992), (Olfati-Saber 2005), (Gan & Harris 2001), (Gao et al. 2009). Distributed state estimation in the case of non-Gaussian models has been studied in (Rosencrantz et al. 2003) where decentralized sensor fusion with the use of distributed particle filters has been proposed in several other research works (Mahler 2007), (Makarenko & Durrant-Whyte 2006), (Deming & Perlovsky 2007). In this paper autonomous navigation of UAVs will be examined and a solution to this problem will be first attempted with the use of the Extended Information Filter and the Unscented Kalman filter (Shima et al. 2007), (Lee et al. 2008), (Lee et al. 2008), (Vercauteren & Wang 2005). Comparatively, autonomous UAV navigation with the use of the Distributed Particle Filter will be studied. This problem belongs to the wider area of multi-source multitarget tracking (Coue et al. 2006), (Hue et al. 2002), (Ing & Coates 2005), (Coue et al. 2003), (Morelande & D. Musicki 2005). Subproblems to be solved for succeeding autonomous navigation of the UAVs are: (i) implementation of sensor fusion with the use of distributed filtering. In this approach the goal is to consistently combine the local particle distribution with the communicated particle distribution coming from particle filters running on nearby