Risk-averse sensor planning using distributed policy gradient

Wann-Jiun Ma, Darinka Dentcheva, Michael M. Zavlanos

Abstract: This paper considers a risk-averse approach to planning the motion of mobile sensor networks in order to maximize the information they collect in uncertain environments. Recent models of risk shape the tails of the probability distributions of the decision variables, controlling in this way the occurrence of rare but important events. In this paper, we formulate the sensor planning problem as a Markov Decision Process (MDP) and propose a distributed risk-averse policy gradient method to obtain optimal policies for the team of sensors. These policies avoid extremely low reward and high risk events. The simulation results validate the effectiveness of the proposed distributed risk-averse method.