A multi-agent reinforcement learning based routing protocol for wireless sensor networks

In this paper, we present MRL-QRP, a multi-agent reinforcement learning based routing protocol with QoS support for wireless sensor networks. In MRL-QRP, sensor node cooperatively computes QoS routes using a distributed value function - distributed reinforcement learning algorithm (DVFDRL). Global optimization can be achieved by using locally observed network information and limited exchanging of state values with immediate neighboring nodes. We compare the network performance of MRL-QRP with QoS-AODV, an on demand QoS support routing protocol. The impact of network traffic load and sensor node¿s mobility on the network performance are investigated, simulation results show that MRL-QRP performs well in respects of a number of QoS metrics and fits well in highly dynamic environments.