Industry 5.0 envisions a cyber-physical future where humans and robots collaborate harmoniously, empowered by 6G connectivity and intelligent automation. Central to this vision is the ability to autonomously configure complex production pipelines based on diverse and evolving human intents. Existing orchestration technologies exhibit critical shortcomings in terms of self-learning, validation, error diagnosis, and rectification capabilities. To this end, we propose an Agentic AI orchestration framework that interprets human intents and dynamically assembles optimal technology pipelines using a self-improving, retrieval-augmented Large Language Model (LLM) and a Bayesian contextual-bandit selector. This enables dynamic adaptation in unpredictable factory environments. Our solution is validated in a cyber-physical testbed integrating Digital Twins (DTs), distributed AI, robotics, and real-world network infrastructure. Compared to baseline LLMs, our system reduces orchestration iterations by over 94% for a given intent and by around 90% for an unseen intent, showing rapid convergence and strong generalization. Real-world deployments mirror DT results, confirming both the fidelity of the simulation and the practical value of intent-driven orchestration for human-centric manufacturing.

The upcoming Sixth Generation (6G) of wireless networks builds on decades of advances and research in scientific areas such as Physics, Electronics & Communication, and Computational Systems. 6G will demand not only ultra-high throughput and low latency, but also scalable, privacy-preserving, and trustworthy coordination across distributed systems. Paradigms such as Fog/Edge Computing (FC/EC), Blockchain (BC), and Federated Learning (FL) each offer innovative solutions to different aspects of these requirements. The combined potential of these technologies, and how they can jointly impact 6G and its applications, are still open issues. Therefore, establishing a roadmap to understand their integration and synergy is needed. In this paper, we present a taxonomy-driven survey. We first analyze FC/EC, BC, and FL and their integration for 6G. Then, as our main contribution, we introduce a comprehensive taxonomy that categorizes seven integration levels (i.e., architecture features, functionalities, security and privacy mechanisms, data management strategies, energy efficiency mechanisms, applications and cross-cutting issues), and explore the effective integration of these paradigms by analyzing their unique characteristics, potential synergies, and opportunities to deliver 6G demands. Through this taxonomy, the study emphasizes transformative benefits such as enhanced data security, improved processing efficiency, and streamlined decentralized privacy-preserving mechanisms, which are actual needs of the in-developing 6G networks. Nonetheless, challenges such as scalability and performance issues and regulatory hurdles are also noted. The work also identifies key areas for future research, and current challenges, promoting exploration into the promising potential of combined FC/EC, BC, and FL solutions for 6G.

Distributed quantum computing (DQC) has emerged as a promising approach to overcome the scalability limitations of monolithic quantum processors in terms of computational capability. However, realising the full potential of DQC requires effective resource management and circuit scheduling. This involves efficiently assigning each circuit to a subset of quantum processing units (QPUs), based on factors such as their computational power and connectivity. In heterogeneous DQC networks with arbitrary connectivity topologies and non-identical QPUs, this becomes a complex challenge. This paper addresses resource management and circuit scheduling in such settings, with a focus on computing resource allocation in a quantum data centre. We propose circuit scheduling algorithms based on Mixed-Integer Linear Programming (MILP). Our MILP model accounts for errors arising from inter-QPU communication. In particular, the proposed schemes consider key factors, including network topology, QPU capacities, and quantum circuit structure, to make efficient scheduling and allocation decisions. Simulation results demonstrate that our proposed algorithms significantly improve circuit execution time and scheduling efficiency (measured by makespan and throughput), while also reducing inter-QPU communication overhead, compared to baseline strategies. This work provides valuable insights into resource management strategies for scalable and heterogeneous DQC systems.

Network slicing (NS) is a promising solution for media services, such as live streaming in telecom networks. NS enables customised network conditions for different applications and requirements. This customisation granularity is further enhanced through the use of 5G Quality of Service (QoS) flows for intra-slice management. Given the fact that network slices are becoming more dedicated to specific services' quality requirements, it becomes important to find service-aware NS methods that guarantee service quality while minimising resource consumption. To this end, this paper presents a Maximum Likelihood-based Network Slicing (MaxLiNS) method that minimises resource consumption with guaranteed Quality of Experience (QoE) for live low-latency streaming service under playback buffer level estimation and control. We evaluated the MaxLiNS method against existing service-agnostic and emerging service-aware NS methods on our End-to-End (E2E) 5G testbed. The results show that the MaxLiNS outperforms existing NS methods in terms of quality guarantee and resource consumption minimisation.

The disaggregated architecture of advanced Radio Access Networks (RANs) with diverse X-haul latencies, in conjunction with resource-limited multi-access edge computing networks, presents significant challenges in designing a general model in placing baseband and user plane functions to accommodate versatile 5G services. This paper proposes a novel approach, NetMind+, which leverages Deep Reinforcement Learning (DRL) to determine the function placement strategies in diverse and evolving RAN topologies, aiming at minimizing power consumption. NetMind+ resolves the problem with a maze-solving strategy, enabling a Markov Decision Process with standardized action space scales across different networks. Additionally, a Graph Convolutional Network (GCN) based encoding and an incremental learning mechanism are introduced, allowing features from different and dynamic networks to be aggregated into a single DRL agent. This facilitates the generalization capability of DRL and minimizes the negative retraining impact. In an example with three sub-networks, NetMind+ demonstrates a substantial 32.76% improvement in power savings and a 41.67% increase in service stability compared to benchmarks from the existing literature. Compared to traditional methods necessitating a dedicated DRL agent for each network, NetMind+ attains comparable performance with 70% of the training cost savings. Furthermore, it demonstrates robust adaptability during network variations, accelerating training speed by 50%.

The emergence of beyond 5G and 6G networks is set to revolutionise telecommunications, addressing the demands of emerging applications through advanced capabilities. At the core of this transformation lies next-generation intelligent service orchestration, which is essential for meeting future Key Performance Indicators (KPIs) and Key Value Indicators (KVIs) such as ultra-low latency, efficient power consumption and resource utilisation. These capabilities require multi-objective, seamless end-to-end service delivery across complex, distributed environments. Achieving such delivery requires scalable and modular system design approaches that support dynamic service composition and adaptability. Cloud-native technologies, underpinned by microservices architectures, plays a pivotal role, but also will introduce challenges in orchestrating resources efficiently across heterogeneous domains. To address these challenges, this paper proposes a solution, Federated Intelligent multi-objective Service function chain Orchestration (FISO) that integrates multi-objective federated profiling to preserve privacy while ensuring efficient end-to-end service delivery. FISO integrates Federated Learning (FL) and Reinforcement Learning (RL). FL is used to collaboratively learn from distributed edge profiling clients without sharing raw data, while RL dynamically guides optimal decision making for resource allocation and Service Function Chain (SFC) placement based on feedback from the federated models. FISO predicts optimal computing and network resources for SFCs, enabling the selection of appropriate edge locations, efficient resource allocation, placement of SFCs, and lifecycle management. Experimental results demonstrated on a pragmatic testbed validate the effectiveness of FISO in efficiently placing requested SFCs within an administrative domain with multiple edge/cloud nodes, predicting optimal CPU, memory, and link capacity resources, and minimising end-to-end latency and energy consumption.

The increasing concerns of knowledge transfer and data privacy challenge the traditional gather-and-analyse paradigm in networks. Specifically, the intelligent orchestration of Virtual Network Functions (VNFs) requires understanding and profiling the resource consumption. However, profiling all kinds of VNFs is time-consuming. It is important to consider transferring the well-profiled VNF knowledge to other lack-profiled VNF types while keeping data private. To this end, this paper proposes a Federated Transfer Component Analysis (FTCA) method between the source and target VNFs. FTCA first trains Generative Adversarial Networks (GANs) based on the source VNF profiling data, and the trained GANs model is sent to the target VNF domain. Then, FTCA realizes federated domain adaptation by using the generated source VNF data and less target VNF profiling data, while keeping the raw data locally. The proposed FTCA enables efficient profiling knowledge transfer among different VNFs, while maintaining data privacy. Through FTCA, faster new VNF deployment can be expected. Experiments show that the proposed FTCA can effectively predict the required resources for the target VNF. Specifically, the RMSE index of the regression model decreases by 38.5% and the R-squared metric advances up to 68.6%.

The adoption of Artificial Intelligence (AI) based Virtual Network Functions (VNFs) has witnessed significant growth, posing a critical challenge in orchestrating AI models within next-generation 6G networks. Finding optimal AI model placement is significantly more challenging than placing traditional software-based VNFs, due to the introduction of numerous uncertain factors by AI models, such as varying computing resource consumption, dynamic storage requirements, and changing model performance. To address the AI model placement problem under uncertainties, this paper presents a novel approach employing a sequence-to-sequence (S2S) neural network which considers uncertainty estimations. The S2S model, characterized by its encoding-decoding architecture, is designed to take the service chain with a number of AI models as input and produce the corresponding placement of each AI model. To address the introduced uncertainties, our methodology incorporates the orthonormal certificate module for uncertainty estimation and utilizes fuzzy logic for uncertainty representation, thereby enhancing the capabilities of the S2S model. Experiments demonstrate that the proposed method achieves competitive results across diverse AI model profiles, network environments, and service chain requests.

Federated Learning (FL) enables collaborative model training across edge devices while preserving data locally. Deploying FL faces challenges due to device heterogeneity. Using cloud technologies like Kubernetes (K8s) can offer computational elasticity, yet may compromise FL privacy principles. K8s can jeopardise FL privacy by potentially allowing malicious FL clients to access other resources given its flat networking approach. This paper introduces the privacy-preserving K8s operator kubeFlower. It addresses privacy risks via isolation-by-design and differential privacy for data management. Isolation ensures secure resource sharing, while differential privacy safeguards individual data privacy. We introduce the Privacy Preserving Persistent Volume Claimer (P3-VC), which adds noise to data while managing a privacy budget. kubeFlower simplifies FL system management in K8s while ensuring privacy. We tested our approach on a network testbed composed of different geo-located cloud and edge nodes where FL clients are deployed. Our results demonstrate the approach’s efficacy in preserving privacy in K8s-based FL for cloud–edge environments.

Abstract Modern software systems are increasingly expected to show higher degrees of autonomy and self-management to cope with uncertain and diverse situations. As a consequence, autonomous systems can exhibit unexpected and surprising behaviours. This is exacerbated due to the ubiquity and complexity of Artificial Intelligence (AI)-based systems. This is the case of Reinforcement Learning (RL), where autonomous agents learn through trial-and-error how to find good solutions to a problem. Thus, the underlying decision-making criteria may become opaque to users that interact with the system and who may require explanations about the system’s reasoning. Available work for eXplainable Reinforcement Learning (XRL) offers different trade-offs: e.g. for runtime explanations, the approaches are model-specific or can only analyse results after-the-fact. Different from these approaches, this paper aims to provide an online model-agnostic approach for XRL towards trustworthy and understandable AI. We present ETeMoX, an architecture based on temporal models to keep track of the decision-making processes of RL systems. In cases where the resources are limited (e.g. storage capacity or time to response), the architecture also integrates complex event processing, an event-driven approach, for detecting matches to event patterns that need to be stored, instead of keeping the entire history. The approach is applied to a mobile communications case study that uses RL for its decision-making. In order to test the generalisability of our approach, three variants of the underlying RL algorithms are used: Q-Learning, SARSA and DQN. The encouraging results show that using the proposed configurable architecture, RL developers are able to obtain explanations about the evolution of a metric, relationships between metrics, and were able to track situations of interest happening over time windows.

Models are not static entities: they evolve over time due to changes. Changes may inadvertently and surprisingly violate constraints imposed. Therefore, the models need to be monitored for compliance. On the one hand, in traditional design-time applications, new and evolving requirements impose changes on a model over time. These changes may accidentally break design rules. Further, the growing complexity of the models may need to be tracked for manageability. On the other hand, newer applications use models at runtime; building runtime abstractions that are used to control a system. Adopters of these approaches will need to query the history of the system to check if the models evolved as expected, or to find out the reasons for a particular behavior. Changes over models at runtime are more frequent than changes over design models. To cover these demands, we argue that a flexible and scalable approach for querying the history of the models is needed to study the evolution and for compliance sake. This paper presents a set of extensions to a model query language inspired in the Object Constraint Language (the Epsilon Object Language) for traversing the history of a model, and for making temporal assertions that will allow the elicitation of historic information. As querying long histories may be costly, the paper presents an approach that annotates versions of interest as they are observed, in order to provide efficient recalls in possible future queries. The approach has been implemented in a model indexing tool, and is demonstrated through a case study from the autonomous and self-adaptive systems domain.