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 booming of the industrial Internet of Things (IIoT) brings an exponential increase in industrial devices, calling for more flexible and low-cost communications. The fifth generation and beyond (B5G) communication technologies provide a dedicated solution by supporting two industry-targeted technologies: Massive machine-type communications (mMTC) and ultra reliable low-latency communications (URLLC). In this article, we design a B5G-aided quality test system in advanced manufacturing, where various sensors are connected to the base station (BS) and send contextual information via mMTC. The BS and quality test machine transmit short length commands and small size feedback to each other, respectively, via URLLC. We formulate a long-term optimization problem to improve the product qualification rate by maximizing the expected average reward with limited testing capacity and changing configurations. To address this problem, we develop a novel context-aware combinatorial quality test (CC-QT) algorithm based on bandit learning (BL), which integrates contextual information to predict the product quality, and a combinatorial method to decrease the complexity of the BL process. Furthermore, we derive a performance upper bound of the proposed CC-QT and analyze its computational complexity. Experimental results illustrate the performance of CC-QT and substantiate its superiority over the existing algorithms.

Integrating mobile edge computing (MEC) in the ultra-dense network (UDN) is a key enabler to meet the service demand by allowing smart devices to perform uninterrupted task offloading via densely deployed MEC servers. In most cases, the smart devices randomly move around the whole network. Consequently, the popular ‘`MEC-centralized decision’' offloading approach could be inapplicable, as joint decision-making among multiple MEC servers becomes difficult due to time synchronization and information exchange overhead. In this paper, we take a user-centric approach to minimize a long-term delay for a given task duration under a price budget constraint. To address this problem, we develop a novel contextual sleeping bandit learning (CSBL) algorithm, which integrates contextual information and sleeping characteristic to accelerate the learning convergence and leverage Lyapunov optimization to deal with the price budget constraint. Furthermore, we extend to a multiple offloading scenario where multiple MEC servers can be selected in each offloading round and propose a CSBL-multiple (CSBL-M) algorithm to address the exponential increase of the offloading selections. For both CSBL and CSBL-M, we derive the upper bounds of learning regret and provide rigorous proofs that they asymptotically approach the Oracle algorithm within bounded deviations for finite task duration.

The industrial Internet of Things (IIoT) interconnects an exponential number of industrial devices, and more flexible and low-cost communications are widely in demand. The fifth-generation (5G) communication provides two industrial-target technologies, massive machine-type communications (mMTC) and ultra-reliable low-latency communications (URLLC), to meet the demand. We design a 5G-aided quality test system, where various sensors are connected to the base station (BS) and send contextual information via mMTC. The BS and quality test machine transmit short-length commands and small-size feedback to each other via URLLC. The problem is formulated as a long-term optimization one with the purpose of improving the product qualification rate. We develop a novel contextual combinatorial quality test (CC-QT) algorithm to solve the problem. We further derive a performance upper bound of the proposed CC-QT and analyze its computational complexity. Experimental results illustrate the performance of CC-QT and substantiate its superiority over the existing algorithms.

The integration of mobile edge computing (MEC) into an ultra-dense network (UDN) can provide ubiquitous task offloading services to computation-demanding users leveraging densely deployed micro base stations. The conventional multi-user task offloading strategies are performed centrally, where a central node makes global task offloading decisions on server selection and resource allocation. In practice, the deployment becomes prohibitively complex with the increasing number of users as it involves high communication overhead and complex global optimization operations. In this paper, we develop a novel decentralized task offloading strategy in UDN, enabling users to independently make local task offloading decisions. We formulate the associated optimization problem to minimize the long-term average task delay among all users. On this basis, we develop a novel calibrated contextual bandit learning (CCBL) algorithm, where users can learn the computational delay functions of micro base stations and predict the task offloading decisions of other users in a decentralized manner. The convergence of the proposed CCBL algorithm is verified via the approachability theory. Moreover, we transfer the target of calibrated learning from all micro base stations to a single user and propose a user-oriented CCBL algorithm to further decrease the computational complexity and increase the convergence rate. Simulation results illustrate that our proposed algorithm outperforms the existing decentralized algorithms and approaches the centralized one.

The rapid development of 5G and Internet-of-Things catalyzes ever-increasing computation-intensive and delay-sensitive applications demanding ubiquitous computation services. Integrating mobile edge computing (MEC) in the ultra-dense network (UDN) is a key enabler to meet the service demand by allowing smart devices to perform uninterrupted task offloading via densely deployed MEC servers. In this paper, we take a user-oriented approach to minimize a long-term delay for a given task duration under a price budget constraint. To address this problem, we develop a novel contextual sleeping bandit learning (CSBL) algorithm, which integrates context information and sleeping bandit theory to handle the fast changing environment and leverages Lyapunov optimization to deal with the price budget. We derive the upper bound of learning regret and provide a rigorous proof that CSBL asymptotically approaches the Oracle algorithm within bounded deviations for finite task duration. Simulation results illustrate that CSBL significantly outperforms existing algorithms.