AIORA: AI-Native 6G Orchestration Architecture

• AIORA is an advanced architectural paradigm that integrates AI-native mechanisms, multi-segment virtualization, and intent-driven APIs for zero-touch orchestration in 6G systems.
• It employs a five-plane logical framework and hierarchical closed-loop controllers to optimize latency, reliability, and resource utilization across heterogeneous domains.
• Standardized with 3GPP, ETSI, and GSMA guidelines, AIORA ensures scalable, resilient, and federated service delivery for ultra-low-latency XR, tactile internet, and V2X applications.

AIORA (AI-Native Multi-Stakeholder Orchestration Architecture) is an advanced architectural paradigm for the management and orchestration of edge-cloud continuum resources in 6G systems. It introduces comprehensive AI-native mechanisms, multi-segment virtualization, and modular, intent-driven APIs to achieve intelligent, zero-touch, and federated orchestration across heterogeneous, multi-domain environments. Building on ongoing 3GPP, ETSI, and GSMA standardization, AIORA targets robust, flexible, and efficient service provisioning—addressing stringent needs for latency, reliability, resource efficiency, and cross-operator federation in future 6G scenarios (Molner et al., 5 Dec 2025).

1. Logical Planes and Core Modules

AIORA is structured as a five-plane logical stack, each featuring specific modules and exposed via open, intent-based APIs:

• User (Northbound) Plane:

Comprises API Gateways implementing CAMARA-style northbound APIs (compatible with 3GPP CAPIF/SEAL) for service subscription, QoS intents, and location queries. The Developer Portal catalogs published virtual continuums, exposes network capabilities, SLA templates, and sandbox environments.

• Intelligence Plane:

Incorporates an AI/ML Pipeline of containerized models (reinforcement learning, transformers, GNNs, and generative approaches) consuming telemetry, digital-twin state, and orchestration logs. The Closed-Loop Coordinator orchestrates nested control loops across radio, compute, and network domains.

• Control (Orchestration) Plane:

Features the AI-Native Orchestration Core, implementing hierarchical decision engines (local orchestrators per virtual continuum, cross-segment resource orchestrator, and federated virtual resource broker), and a Life Cycle Manager for multi-segment virtual continuums and 3GPP SA6 edge functions (EAS, EES, ECS).

• Data (Infrastructure) Plane:

Consists of the Ubiquitous Computing Manager abstracting diverse Kubernetes clusters, VMs, and containers, and the Infrastructure Configuration & Resource Manager managing instantiation of RAN, 6G-Core functions, MEC platforms, network slice configuration, and security domains.

• Management Plane:

Encompasses Cross-Segment Intelligence & Resource Coordination for digital twin ingestion, conflict detection, and automated/human-in-the-loop negotiation. The Telemetry & Digital Twin Substrate aggregates real-time KPIs, topology, user mobility, energy and CO₂ data, feeding both AI pipelines and closed-loop controllers.

<table> <thead> <tr> <th>Plane</th> <th>Key Modules</th> <th>Interfacing Technologies</th> </tr> </thead> <tbody> <tr> <td>User</td> <td>API Gateway, Developer Portal</td> <td>CAMARA, 3GPP CAPIF/SEAL</td> </tr> <tr> <td>Intelligence</td> <td>AI/ML Pipeline, Closed-Loop Coordinator</td> <td>Telemetry, Digital Twin State</td> </tr> <tr> <td>Control</td> <td>Orchestration Core, Lifecycle Manager</td> <td>Intent-based APIs</td> </tr> <tr> <td>Data</td> <td>Ubiquitous Computing Manager</td> <td>K8s, VM, MEC</td> </tr> <tr> <td>Management</td> <td>Resource Coordination, Telemetry Substrate</td> <td>OpenTelemetry, Digital Twins</td> </tr> </tbody> </table>

2. Multi-Segment Virtual Continuum

The “multi-segment virtual continuum” (MSVC) is AIORA’s central abstraction, managing federated resource pools across edge, cloud, RAN, and network core, even when spanning multiple administrative or operator domains. An MSVC is defined as a configurable set of:

• RAN nodes (gNB-CU/DU/O-RU)
• Core network functions (UPF, SMF, AMF)
• Compute/storage clusters (Kubernetes, VMs, bare-metal)
• Network slices, transport tunnels, security zones
• AI-native control loops

Each MSVC is parameterized to meet specific application profiles such as latency, reliability, mobility, data sovereignty, and energy budgets.

Unification mechanisms include:

• The Virtual Resource Broker, which leverages federated catalogs and SLA-based offerings from multiple operators (GSMA Operator Platform) or domains (ETSI MEC).
• Digital Twin Federation, aggregating per-domain twins into cross-domain digital twins for capacity planning and resource conflict detection.
• High-level intent translation, allowing service requirements (e.g., “4 ms RTT, 99.999% reliability for XR”) to be decomposed into discrete orchestrator actions—placement, slicing, RAN configuration—by the Orchestration Core.

3. AI-Native Nested Closed-Loop Control

AIORA’s orchestration relies on deep integration of hierarchical, nested closed-loop controllers at multiple system layers:

• Intra-Segment (Local) Loops:
  • Radio Loop: Per gNB cluster, leveraging channel state, UE location, and traffic demand to maximize throughput $T(t)$ and minimize interference. Example Q-learning update law:

  $$Q_{t+1}(s,a) = Q_t(s,a) + \alpha [r_t + \gamma \max_{a'} Q_t(s',a') - Q_t(s,a)]$$ - Compute Loop: For each Kubernetes cluster, minimizing

  $\alpha L_\text{compute}(t) + \beta E(t)$

  via gradient descent on a surrogate performance model:

  $$\theta_{t+1} = \theta_t - \eta \nabla_\theta [\alpha L_\text{compute}(\theta) + \beta E(\theta)]$$ - Network Loop: Per slice, enforcing $RTT \leq RTT_{\max}$ and $p_\text{loss} \leq p_{\max}$, adjusting queue weights and paths through a learned or closed-form controller.

• Cross-Segment (Global) Loop:

A top-level RL agent operates over the global state $$s_\text{global}(t) = \{T_\text{radio},L_\text{compute}, RTT, \text{energy}_\text{tot}\}$$, jointly optimizing placement, slice configuration, and RAN parameters according to:

$$\min J(x) = \sum_{i=1}^S [\alpha L_i(x) + \beta E_i(x) - \gamma U_i(x)]$$

subject to resource, latency, and reliability constraints.

• Closed-Loop Coordination:

The Closed-Loop Coordinator arbitrates between local and global objectives, resolving actions via weighted utility maximization:

$$a^* = \arg\max_{a \in \{a_\text{global}, a_\text{local}\}} w_g U_\text{global}(a) + w_l U_\text{local}(a)$$

4. Integration with 3GPP, ETSI, and Zero-Touch Mechanisms

AIORA is designed atop industry-standard 3GPP and ETSI platforms:

• 3GPP SA6 TS 23.558 EDGEAPP:

The EAS/EES/ECS logical functions directly map to AIORA’s service enabler layer.

• 3GPP CAPIF/SEAL Northbound APIs:

Evolving APIs expose unified network capabilities, feeding resource, event, and location data directly into the orchestration stack.

• 3GPP Connectivity Models:

User-plane function topologies, NPN/NSSI instances, and dynamic coverage maps are ingestible as digital twin state representations.

• ETSI ZSM Zero-Touch & ENI:

Closed-loop taxonomies, ML-guided control, and experiential intelligence enable full zero-touch service management throughout continuum operations.

5. Standardized Open APIs and Federation Models

AIORA’s multi-stakeholder focus is realized through model-driven, secure, and interoperable APIs:

• ETSI MEC:

MEC Platform Manager and Service APIs provide modular integration for per-continuum edge orchestration.

• GSMA Operator Platform:

Implements standardized federation—including listing and leasing cross-operator edge and cloud resources.

• CAMARA (3GPP CAPIF enhancement):

Northbound APIs for network-aware, slice-level programmability are provided to third-party developers.

• O-RAN RIC Integration:

Interfacing with the O-RAN non-RT RIC, including rApps/mApps, enables direct feedback between orchestration core and RAN policy mechanisms.

APIs are YANG/TOSCA-modelled and secured with CAPIF OAuth2.0 flows, supporting true multi-tenant, multi-stakeholder, zero-touch provisioning, and federation.

6. Performance and Optimization Frameworks

AIORA supports a range of quantitative optimization models beyond its primary multi-objective RL loops:

• Service-Chain Placement:

$$\min J_\text{place} = \sum_{s=1}^S \sum_{r=1}^R x_{s,r} (w_1 L_{s,r} + w_2 C_{s,r})$$

with resource capacity, placement, and latency constraints.

• RAN Parameter Optimization:

$\max T = \sum_{u=1}^U \log(1 + \text{SINR}_u(P))$

subject to transmission and interference limits.

• Energy-Aware Compute Scaling:

$\min E = \sum_{k=1}^K P_k(n_k)$

under workload and resource activation constraints.

Telemetry streams supply key performance metrics: end-to-end latency, slice reliability ($1 - p_\text{loss}$), CPU/GPU utilization, energy per bit, and migration downtime.

7. Representative Application Scenarios

AIORA’s mechanisms are exemplified in use cases characterized by extreme requirements for latency, reliability, and adaptability:

• Ultra-Low-Latency XR Streaming:

AIORA dynamically constructs MSVCs comprising geographically proximate micro-data centers, XR-specific UPF slices, and GPU-enabled K8s pods, with nested RL-based loops tuning RAN and compute resources within 10 ms to maintain 4 ms RTT and 99.999% reliability.

• Tactile Internet in Manufacturing:

Federated orchestration across private 6G NPNs and public edge clouds uses digital twin prediction to pre-migrate control applications, maintaining ≤1 ms control feedback without operator intervention.

• Resilient V2X Handover:

Orchestration of RAN reconfiguration, container migration, and path re-optimization in response to vehicular mobility ensures $PDR > 99.99\%$ and handover latencies below 50 ms.

Across scenarios, AIORA’s closed-loop architecture maintains real-time SLA assurance and system resilience with full automation and zero human intervention.

---

AIORA converges multi-segment virtualization, AI-driven orchestration, and open federation into an autonomous, resilient 6G continuum. It extends prevailing standards—3GPP SA6, ETSI MEC/ZSM, GSMA Operator Platform, and CAMARA—by integrating generative AI for system configuration and cross-domain digital twin coordination, delivering real-time, multi-objective optimization for next-generation services (Molner et al., 5 Dec 2025).