Researchers introduce a distributed filtering algorithm that helps networks of sensors track a moving target even when communication links between sensors change over time. The Dynamic Generalized Kalman Consensus Filter (DGKCF) avoids the need for global network information, so each sensor can form better estimates using only local data and messages from neighbors.

The paper addresses a common problem in distributed tracking. Classic approaches like the Kalman Consensus Filter (KCF) average information from neighbors equally. That equal weighting becomes a problem when some sensors are “oblivious” — they have no current direct measurements and may pass along out-of-date estimates. The issue gets worse if sensors move and links switch on and off, because many existing methods need global graph parameters (for example, the maximum number of neighbors any node has) that change with the topology.

To fix this, the authors develop the DGKCF. The new algorithm embeds a weighted consensus step inside a Kalman-style update. Each agent computes consensus weights from information it already has locally, rather than using a network-wide constant. In plain terms, sensors give more credence to neighbors whose information looks more reliable and less to those that appear oblivious, and they do this without learning the whole network structure.

Why this matters: numerical simulations in the paper show that DGKCF keeps accurate tracking when the communication topology switches and when some agents lack measurements. In the tested scenarios the method produced lower root mean square error (RMSE), lower mean absolute error (MAE), and faster convergence than several other distributed filters. The authors also report that the computational cost is comparable to other distributed filters that use weighted consensus.