We consider a dynamic network of sensors that cooperate to estimate parameters of multiple targets. Each sensor can observe parameters of a few targets, reconstructing the trajectories of the remaining targets via interactions with “neighbouring” sensors. The multitarget tracking has to be provided in the face of uncertainties, which include unknown-but-bounded drift of parameters, noise in observations and distortions introduced by communication channels. To provide tracking in presence of these uncertainties, we employ a distributed algorithm, being an “offspring” of a consensus protocol and the stochastic gradient descent. The mathematical results on the algorithm's convergence are illustrated by numerical simulations.