Abstract
Quality of Service (QoS) assurance is critical for maintaining reliable communication in post-disaster environments where fixed infrastructure is often damaged or unavailable. In such scenarios, routing must rely on mobile and ad hoc nodes that face frequent topology changes and unstable links. Conventional routing protocols such as AODV, OLSR, and GPSR depend on static metrics and are unable to adapt to these dynamic conditions. This paper presents a reinforcement learning-based routing framework built on the Soft Actor Critic (SAC) algorithm to enable adaptive QoS-aware communication. The model observes network states, including link delay, packet reliability, and node energy, and learns routing policies that maintain stable multi-hop connectivity under varying mobility. The framework is evaluated in a simulated post-disaster network that includes mobile edge devices, UAV relays, and temporary gateways. Results show that the SAC-based model reduces average round-trip time by about 18% compared with PPO and achieves nearly 95% of the throughput upper bound under static conditions. When 60% of nodes are mobile, it retains over 90% of baseline throughput and maintains a stable success rate. These findings suggest that SAC offers a robust and adaptive approach for QoS-driven routing in dynamic networks.