The National Academies Logo
Research Associateship Programs
Fellowships Office
Policy and Global Affairs

Participating Agencies - AFRL

  Sign InPrintable View

Opportunity at Air Force Research Laboratory (AFRL)

Robust, Self-tuning Estimators in Distributed Systems

Location

Sensors Directorate, RY/Electromagnetics Technology Division

RO# Location
13.35.01.B7606 Wright-Patterson AFB, OH 454337542

Advisers

Name E-mail Phone
Taylor, Clark Noble clark.taylor.3@us.af.mil 937.713.8184

Description

In a simplistic sensor fusion environment, each sensor generates data, together with an uncertainty estimate of that data. Then, when sensor fusion is to occur, the uncertainties on each sensor are used to weight each sensor measurement according to well-known estimation algorithms such as the Kalman filter. The purpose of this topic is to explore (individually or jointly) four main difficulties associated with this general framework. First, in many cases the uncertainty due to a certain set of measurements may not be known a-priori. This may be due to the expense of calibration procedures, or due to the dependence of the uncertainty on operatings conditions that cannot all be specified a-priori. Therefore, estimation algorithms that can perform accurate data fusion, even without accurate a-priori measures of uncertainty, are of interest to this topic. Second, once a model for uncertainty is known, outliers (rare events that do not fit the uncertainty model) may still be present in the data. Ensuring the estimation algorithm is robust to outliers, while still generating accurate estimates of uncertainty for its a-posterior estimates are another area of interest. Third, models for uncertainty, especially when the uncertainty has been estimated from samples (a possible approach to solving the first problem), can be very complex and may not even be have a closed-form representation. Performing Bayesian fusion on these general (e.g., non-parametric) representations of uncertainty then becomes even more difficult. Advances in implementing fusion of different, sampled distributions are therefore desired to enable more complex estimation routines. Fourth, performing estimation is a fully distributed system, where there is limited communication between sensors/agents, but better estimation of system state can be achieved using data from all agents, requires new algorithms and approaches than outlined in the "simplistic" scenario above. Any of these four, estimation theory based topics are of significant interest to this topic.

 

Keywords:
Estimation theory; Outlier rejection; Distributed data fusion; Sampling theory; Bayesian estimation; Uncertainty quantification;

Eligibility

Citizenship:  Open to U.S. citizens
Level:  Open to Postdoctoral and Senior applicants
Copyright © 2014. National Academy of Sciences. All rights reserved. 500 Fifth St. N.W., Washington, D.C. 20001.
Terms of Use and Privacy Statement.