Stochastic Systems Group
Home Research Group Members Programs  
Demos Calendar Publications Mission Statement Alumni

SSG Seminar Abstract


Hybrid State Multiple Model Particle Filtering from Sparse Sensor Measurements


Emily Fox - SSG, MIT


In the field of biological sensing, quick and accurate detection of a biological attack is required from a sparse set of noisy sensor measurements. Because particle flow dynamics can be approximately modeled, a Bayesian filtering formulation can provide information about the entire region from the sequence of uncertain measurements. While not required for detection alone, the knowledge of system dynamics may enable localization of a release in both time and space. The formulation also constrains the set of solutions examined in the pure detection problem as well as providing a framework in which to analyze performance as a function of sensor density, wind field uncertainty, etc.

Due to the fact that the entire region is not fully observable, we propose a method which formulates a set of hypotheses of potential releases at every time step based on a Hidden Markov Model (HMM). A given discrete hypothesis dictates the dynamics for the associated continuous state. The effects of the hypotheses persist over time which allows us to characterize how well our observations match the estimated values associated with each hypothesis at every time step. The talk primarily focuses on the formulation of the hybrid state multiple model particle filter, though some preliminary results will be presented.



Problems with this site should be emailed to jonesb@mit.edu