Reinforcement Learning Aided Routing in Tactical Wireless Sensor Networks

A wireless sensor network (WSN) consists of a large number of sensor nodes with limited battery lives that are dispersed geographically to monitor events and gather information from a geographical area. On the other hand, tactical WSNs aremission-critical WSNs that are used to support military operations, such as intrusion detection, battlefield surveillance, and combat monitoring. Such networks are critical to the collection of situational data on a battlefield for timely decision-making. Due to their application area, tactical WSNs have unique challenges, not seen in commercial WSNs, such as being targets for adversarial attacks. These challenges make packet routing in tactical WSNs a daunting task. In this article, we propose a multi-agent Q-learning-based routing scheme for a tactical WSN consisting of static sensors and a mobile sink. Using the proposed routing scheme, a learning agent (i.e., network node) adjusts its routing policy according to the estimates of the Q-values of the available routes via its neighbors. The Q-values capture the quickness, reliability, and energy efficiency of the routes as a function of the number of hops to sink, the one-hop delay, the energy cost of transmission, and the packet loss rate of the neighbors. Simulation results demonstrate that, in comparison to a baseline random hop selection scheme, the proposed scheme reduces the packet loss rate and mean hop delay, and enhances energy efficiency in the presence of jamming attacks.