Collaborative Intelligence: Orchestrating Ultra-Low Latency XR Experiences in 6G Consumer Networks

The rapid evolution of Extended Reality (XR) demands ultra-low latency that centralized cloud architectures cannot reliably deliver. This article proposes a Collaborative Intelligence (CI) framework for 6G-enabled consumer networks, combining Hierarchical Split-Federated Learning with Peer-to-Peer (P2P) knowledge distillation within an Open Radio Access Network (O-RAN) compliant architecture. The RAN Intelligent Controller (RIC) dynamically manages network slices using a formally defined offloading ratio and slice-aware reward function governing multi-agent resource allocation. Co-simulation using ns-3 and PyTorch suggests that CI can maintain end-to-end latency well within the 20 ms XR threshold, while accelerating convergence by approximately 1.75 times through P2P knowledge distillation and reducing per-round training payload over quantized edge-FL baselines. These simulation results suggest CI as a promising enabler for seamless XR deployment in the future consumer metaverse.