This keynote introduces a Distributed Artificial Intelligence (DAI) framework that pushes “brains to the edge” via BDIx agents—BeliefDesireIntention agents extended with priority-aware plan libraries and lightweight fuzzy/ANFIS reasoning, communicating through well-defined messages (e.g., FIPA ACL) to realize fast, low-overhead self-organization and self-healing in 5G/6G. The framework integrates with existing AI/ML modules (RL, forecasting, anomaly detection) while remaining modular and device-centric, enabling autonomy from and collaboration with network infrastructure where applicable (UE-VBS, edge execution). We ground the design in D2D communications across licensed/unlicensed spectrum, enabling agents to coordinate device discovery, mode selection (relay, multihop, clustering), interference/power control, and security. Decisions are guided by a Weighted Data Rate (WDR) objective, while mobility rules prevent unstable relay choices; protocol soundness is validated through formal analysis. In simulations with 10–1000 devices (using Poisson spatial models), the framework increases the sum rate, reduces energy consumption, and reduces signaling compared to centralized or heuristic baselines in both static and dynamic environments (where UEs have speeds greater than 1.5 m/s). Finally, we map BDIxDAI into the ADROIT6G architecture: AI-driven management/orchestration, crowdsourced/federated AI services, and unified BDIx-based control loops spanning cloud–edge–faredge, and illustrate PoCs including holographic teaching, IIoT, NTN-assisted rail, and collaborative robots meeting stringent 6G KPIs (sub-ms latency, 99.999% reliability, 10× energy efficiency). The takeaway: a practical path to trustworthy, transparent, and resilient autonomy edge-first, cloud-connected, and self-organizing by design.

暂无