A Distributed <scp>QoS</scp> ‐Aware <scp>MANET</scp> Routing Protocol Based on Deep Reinforcement Learning and <scp>SDN</scp> Integration

ABSTRACT Mobile ad hoc networks (MANETs) face significant routing challenges in dynamic environments due to frequent topology changes, limited bandwidth, and resource constraints. To address these issues, this paper proposes a novel routing framework that integrates deep reinforcement learning (DRL) with software‐defined networking (SDN) to enable adaptive, QoS‐aware routing. The system models the routing process as a Markov decision process, where a DRL agent employs a gated recurrent unit (GRU)‐based actor–critic architecture to learn optimal link weight adjustments from real‐time network states. A dynamic weight strategy and an action discretization module allow the model to operate efficiently in a continuous action space while maintaining compatibility with SDN‐based control deployment. The routing policy is optimized using a multi‐objective reward function that targets throughput, delay, and jitter. The framework is implemented using Mininet and PyTorch and evaluated against traditional routing protocols (AODV and OLSR) as well as an SDN‐based static routing baseline. Experimental results show that the proposed method improves average throughput by 15.26% over the next best protocol, while simultaneously reducing average delay and jitter by 44.59% and 10.90%, respectively. These findings validate the potential of distributed DRL to provide scalable, intelligent routing solutions in highly dynamic MANET environments.