A Distributed Policy Gradient Algorithm for Optimal Coordination of Mobile Sensor Networks

In this paper, we study the optimal deployment problem in mobile sensor networks. By recasting it as an optimal coordination problem of multiagents, a new distributed policy gradient algorithm is proposed based on the minimization of the overall cost for all agents. The proposed algorithm relies on local information exchanges among neighboring agents without the requirement of known system dynamics. The policy gradient is computed based on sampling the trajectory under the perturbed control policy. The control policy is parameterized and the adaptive parameter update is carried out following the negative gradient of the overall cost. The rigorous analysis of the proposed algorithm is provided.