Hydra-RAN Task 3: A Core-Independent AI Framework for Robust Intra-SRU Switching (ISS) for 6G Networks

Existing 5G mobility schemes fail under severe millimeter-wave blockage due to reactive, radio-only triggers, causing latency spikes and service interruption. This paper introduces Hydra-RAN Task 3, a core-independent AI framework featuring a dual-policy intra-SRU switching (ISS) mechanism. The framework uniquely integrates: (i) a sensor-driven proactive (Detective) policy for anticipatory switching; (ii) a measurement-driven reactive (Reactive) policy for robust fallback; and (iii) a lightweight deep reinforcement learning (DRL)-based selector. A sparse multi-task learning (SMTL) engine enables efficient multi-modal sensing fusion. Comprehensive simulations demonstrate that the proposed architecture achieves up to 75% reduction in blockage recovery time, maintains URLLC-compliant latency (sub-1 ms) in over 92.5% of events, and provides robust out-of-distribution (OOD) generalization (89.7% URLLC compliance versus 47.3% for baselines), representing a 17% improvement in mean accuracy over the best-performing baseline. This work fills a critical gap in 6G literature by presenting a rigorously evaluated, dual-policy, sensor-RF fused ISS architecture.