ABSTRACT Mobile Ad Hoc Networks (MANETs) are inherently susceptible to energy depletion and non-cooperative node behavior, both of which critically degrade routing performance and network lifetime. Existing routing protocols fail to simultaneously address dynamic topology changes, selfish node detection, and energy-balanced path selection. This paper presents REAR-RL (Reinforcement Energy-Aware Routing via Reinforcement Learning), a novel adaptive routing framework that integrates a Q-learning-based decision engine with a multi-criteria reward function encapsulating residual energy, link quality, node cooperation history, and hop count. REAR-RL employs a lightweight trust model derived from packet forwarding behavior to identify and isolate non-cooperative nodes without requiring centralized infrastructure. The reward shaping strategy prioritizes routes that balance energy consumption across participating nodes while maximizing packet delivery. Extensive simulations conducted in NS-3 with 50 to 200 mobile nodes reveal that REAR-RL achieves up to 34.7% improvement in network lifetime, a 28.3% increase in packet delivery ratio, and reduces end-to-end delay by 19.6% compared to AODV, DSR, and OLSR under varying node mobility and traffic loads. These results demonstrate the viability of model-free reinforcement learning as a scalable, infrastructure-free solution for intelligent routing in adversarial mobile environments.
Chimdesa Gedefa (Sun,) studied this question.