Algorithms for Networked Cooperative Navigation

Principal Investigator(s):

Demoz Gebre-Egziabher, Professor/Director, Aerospace Engineering & Mechanics

Project summary:

The work in this proposal deals with developing algorithms used to fuse the relative range/angle measurements from cooperating vehicles with other information shared on the network to form an accurate position solution.

a) Research Question: The focus of this research is the development of decentralized sensor fusion algorithms which 1) are accurate, 2) minimize use of communication bandwidth, and (3) are immune to error-loop formation. An error-loop is a phenomenon which is seen in decentralized estimators and is somewhat analogous to an unstable feedback system. As an illustration, imagine that Member A of the cooperative navigation community shown in Figure 1 receives information from Member B. In this information exchange, any inaccuracies present in Member A's position estimate will be transmitted (in whole or part) to Member B. Subsequently, Member B exchanges information with Member C and Member A's errors now become integrated into Member C's position estimate. If Member A then collaborates with Member C, it will receive information that is corrupted by its own original errors. The end result is that the effect of the original error can increase exponentially, rather than decaying to zero. The research question, therefore, is how to design decentralized sensor fusion algorithms to mitigate the effect of error loops.

(b) Prior Work: It has been demonstrated in prior work that algorithms can be designed to mitigate the effect of error loops. To deal with error loops in decentralized sensor fusion schemes, a class of algorithms known collectively as covariance intersection (CI) filters have been developed. The authors of this proposal have shown how variants of CI can be used to implement cooperative navigation in automotive navigation applications and unmanned aerial vehicle applications. In a previous publication, the authors of this proposal expounded on the theoretical underpinning of CI-based algorithms and identified one of the keys to improving their accuracy. CI-based algorithms mitigate the effect of error loops at the cost of decreased accuracy because they treat all information from a particular member of the network as being of the same quality. If information from another cooperating vehicle is of mixed quality (i.e. some high accuracy, some low accuracy), CI-based algorithms cannot selectively place more emphasis on the accurate information while de-emphasizing the poor information. As such, they will reject all the information as being of poor quality in order to be conservative. This leads to the potentially useful information being rejected, which in turn leads to decreased accuracy. The work proposed here is to develop algorithms which remove some of this unnecessary conservatism.

(c) Proposed Solution: The goal of the work described in this proposal is to develop a mathematically sound formulation of the CI filters that can selectively weigh partial information from a particular member of the network. As has been shown in prior work, this will require formulating the CI-filters with a tuning parameter that is a matrix and not a simple scalar. The algorithms developed will be validated on archived data from flight tests of unmanned aerial vehicles (UAVs) from the University of Minnesota UAV laboratory.

Project details: