The rapid evolution of immersive and bandwidth-intensive applications such as 360-degree video, Virtual Reality (VR), Augmented Reality (AR), Extended Reality (XR), and Volumetric Video (VV) poses substantial challenges to contemporary network infrastructures, especially in delivering high perceptual Quality of Experience (QoE) and real-time content adaptability. These challenges are intensified in Fifth Generation (5G) and Sixth Generation (6G) environments, where heterogeneous user demands, device capabilities, and network conditions necessitate intelligent, context-aware streaming solutions. To address these complexities, this thesis investigates the integration of Artificial Intelligence (AI) techniques with enabling technologies in the context of network slicing, including Software-Defined Networking (SDN), Network Functions Virtualization (NFV), and a specifically proposed Multi-Access Edge Computing (MEC)-based Adaptive Bitrate (ABR) streaming strategy to support rich media delivery. The thesis proposes EDGE360, a novel edge-assisted Multi-Agent Deep Reinforcement Learning (MADRL) framework for region-aware ABR control in tile-based 360-degree video streaming. EDGE360 employs the Counterfactual Multi-Agent (COMA) policy gradient method to optimize bitrate allocation across viewport, marginal, and background regions, using the Video Multimethod Assessment Fusion (VMAF) metric and viewport trajectory predictions. A centralized critic collaborates with distributed agent networks deployed at the edge, leveraging real-time feedback from client buffers, throughput estimations, and tile-level monitors. The proposed solution models the adaptation task as a Partially Observable Markov Decision Process (POMDP) and ensures joint QoE optimization and fairness across multiple clients. Experimental results confirm that EDGE360 significantly outperforms existing state-of-the-art ABR schemes by achieving the highest average reward, outperforming RAPT360, 360SRL, and BOLA360 by 8.12%, 11.86%, and 18.00%, respectively, demonstrating superior convergence and refinement under dynamic network conditions. Extending adaptive streaming to next-generation VV scenarios, the thesis also introduces METAVV, a novel saliency-aware viewport-adaptive Meta-Reinforcement Learning (Meta-RL) framework built on the PEARL (Probabilistic Embeddings for Actor–Critic Reinforcement Learning) paradigm. Formulated as a POMDP, METAVV incorporates 6-DoF viewport prediction, Levels of Detail (LoD) mesh segmentation, and perceptual quality assessment based on the Point Cloud Quality Metric (PCQM). By inferring latent belief vectors from session-level observations, METAVV dynamically conditions the ABR policy for fast adaptation across out-of-distribution network and user mobility scenarios. This context-aware design enables rapid convergence, personalized streaming decisions, and consistent perceptual quality even under constrained or unseen conditions. Evaluations using public VV datasets show that, under near-distance viewing, METAVV improves average QoE by 3.25%, 3.78%, 11.98%, and 18.91% over FewVV, Sparkle, Rolling POT, and ViVo, respectively. Under far-distance viewing, the corresponding QoE gains are 8.81%, 7.17%, 16.94%, and 24.93%, respectively. These results demonstrate that the proposed solution outperforms state-of-the-art solutions in rate adaptation under dynamic and out-of-distribution conditions. The thesis also includes a comprehensive survey that discusses AI-driven advancements for intelligent content-centric networking, predictive streaming, cache optimization, and secure slice-aware traffic management. It highlights how AI models ranging from Machine Learning (ML) to DRL are reshaping slice orchestration and media delivery in future networking environments.