Abstract
Ad hoc networks persistently struggle to guarantee stringent Quality of Service (QoS) when node mobility, interference, and heterogeneous traffic patterns compete for scarce wireless resources. This article proposes an AI-enhanced routing and slicing framework for Mobile Ad Hoc Networks (MANETs) that couples Deep Reinforcement Learning (DRL) with adaptive Network Slicing (NS) to steer packets through latency aware, slice specific paths. The DRL agent observes local topology changes, queue states, and slice budgets, then selects next hops that jointly minimize end to end delay and maximize packet delivery ratio, while a fuzzy logic slicer reallocates bandwidth across slices in real time. We trained the agent in MATLAB using Proximal Policy Optimization and implemented slice control with native Communications System Toolbox functions. Simulations over 100 to 300 nodes moving under the Random Waypoint model showed that, compared with Ad Hoc On Demand Distance Vector (AODV), Dynamic Source Routing (DSR), and a standalone DRL router, the proposed scheme educed average delay by 37%, increased throughput by a factor of 1.8, and lifted packet delivery ratio by 22% at node speeds up to 20 m/s, without sacrificing energy efficiency or incurring excessive control overhead. These results confirm that integrating intelligent routing with agile slicing is a viable pathway to sustain application level QoS in highly dynamic MANETs.