Reinforcement learning based reliability-aware routing in IoT networks

The unprecedented scale and ubiquity of the Internet of Things (IoT) introduce a maintainability challenge. IoT networks operate in diverse and harsh environments that impose thermal stress on IoT devices. The lifetime of these networks can be limited by hardware failures resulting from exacerbated reliability degradation mechanisms at high temperatures. In this paper, we propose a novel adaptive and distributed reliability-aware routing protocol based on reinforcement learning to mitigate the reliability degradation of IoT devices and improve the network Mean Time to Failure (MTTF). Through routing, we curb the utilization of quickly degrading devices, which helps to lower the device power dissipation and temperature, thus reducing the effect of temperature-driven failure mechanisms. To quantify and optimize networking performance besides reliability, we incorporate Expected Transmission Count (ETX) in our formulations as a measure of communication link quality. Our proposed algorithm adapts routing decisions based on the current reliability status of the devices, the amount of degradation they are likely to experience due to communication activity, and network performance goals. We extend the ns-3 network simulator to support our reliability models and evaluate the routing performance by comparing with state-of-the-art approaches. Our results show up to a 73.2% improvement in reliability for various communication data rates and the number of nodes in the network while delivering comparable performance.