Autonomous and intelligently coordinated systems are central to the evolution of the Internet of Things (IoT) and smart industry paradigms.  The envisioned digital transformation in the context of smart network platforms (5G/6G) is expected to establish a connected ecosystem characterized by significant design flexibility, rapid and cost-efficient adaptability, enhanced predictability, and improved controllability.  While the fifth generation (5G) of mobile communication systems initiated the digital transformation in data coordination by introducing flexible, programmable serviceability, 6G aims to fully realize communication technology revolution by incorporating AI, learning capabilities and edge computing. These advancements will enable the smart and task-driven utilization of network resources, which is crucial for providing cost-effective communication services with the quality of service necessary for efficient operations. To this end, it is essential to develop novel, joint resource allocation strategies between the physical components and the network resources, considering both operational and serviceability constraints of the industrial cyber-physical systems. In this context, we study interactive and task-oriented schemes for cyber-physical systems relying upon 5G-and-beyond systems to support the stringent communication requirements of physical entities towards safe and reliable real-time operations.

Performance of control systems interacting over a shared communication network is tightly coupled with how the network provides services and distributes resources.  Novel networking technology such as 5G is capable of providing tailored services for a variety of network demands. Stringent control requirements and their critical performance specification call for online adaptable and control-aware network services. This perspective suggests a co-design of physical and network layers aiming to ensure that the necessary quality-of-service is provided to achieve the desired quality-of-control. An optimal co-design is in general challenging due to cross-layer couplings between the physical and network layers and their layer-specific functionalities. Furthermore, the complexity of the co-design depends on the level of actionable information the layers share with each other. We study general co-design of physical operations and service allocation aiming to minimize a social regret measure for networked control systems. We also investigate optimal networked co-design scenarios using the regret index as the joint quality-of-control and quality-of-service (QoC-QoS) measure, and try to understand the role of cross-layer awareness in the structure of optimization problems. We focus on both the finite-horizon and the infinite-horizon cases. 

A strategic application of Net-CPS that needs radical performance and safety improvements is autonomous mobile systems (robotics & autonomous cars); a safety-critical, multi-modal, and largely interactive scenario with coordinated and non-coordinated interactions among physical entities (vehicles, robots, humans, environment) and cyber entities (perception data, cloud, network, AI). Pattern of interaction is often not fixed or deterministic due to the weakly predictable and stochastic behavior of surrounding's physical entities, dynamic modality of interacting nodes, and failures in communicating, hence, low and middle level safety guarantees are specifically challenging to enforce. This means safe planning and control strategies need to be responsive not only to microscopic behavior of physical nodes, but also to the geometrical and temporal adaptations of the multi-layer interaction graph. 

To perform co-planning and co-design, reliable state estimation and intention prediction is required, while awareness of the operating environment is key. This will form the perception part which collects the required data for policy/decision making. Advanced proprioceptive sensors (IMU, wheel encoders, speedometers) measure internal systems’ states accurately required for reliable inertial navigation, while exteroceptive sensors (Lidar, camera) provide real-time point clouds and map of the environment and moving objects required for real-time navigation. Furthermore, communication with surrounding objects enhances awareness of their intents and situation through V2X communications. All these provide large amount of data in various forms (images, numerical, stream) that entail various reliability, time resolutions, and spatial features. Having various data, next step is to transform data into actionable knowledge and eventually actuation policies. Major research areas include robust and reliable state estimation for localisation and navigation using proprioceptive and exteroceptive sensors; and monitoring & mapping semi-structured and unstructured environments in almost real-time for safe motion planning. Control, performance optimization, and policy adaptation completes the loop of autonomy.