The emergence of Tiny Machine Learning (TinyML) has positively revolutionized the field of Artificial Intelligence by promoting the joint design of resource-constrained IoT hardware devices and their learning-based software architectures. TinyML carries an essential role within the fourth and fifth industrial revolutions in helping societies, economies, and individuals employ effective AI-infused computing technologies (e.g., smart cities, automotive, and medical robotics). Given its multidisciplinary nature, the field of TinyML has been approached from many different angles: this comprehensive survey wishes to provide an up-to-date overview focused on all the learning algorithms within TinyML-based solutions. The survey is based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodological flow, allowing for a systematic and complete literature survey. In particular, firstly, we will examine the three different workflows for implementing a TinyML-based system, i.e., ML-oriented, HW-oriented, and co-design. Secondly, we propose a taxonomy that covers the learning panorama under the TinyML lens, examining in detail the different families of model optimization and design, as well as the state-of-the-art learning techniques. Thirdly, this survey will present the distinct features of hardware devices and software tools that represent the current state-of-the-art for TinyML intelligent edge applications. Finally, we discuss the challenges and future directions.

With the advent of Industrial 4.0 and the push toward Industry 5.0, the data generated by the industries have become surprisingly large. This abundance of data significantly boosts machine and deep learning models for Predictive Maintenance (PdM). The PdM plays a vital role in extending the lifespan of industrial equipment and machines while also helping to reduce the risk of unscheduled downtime. Given its multidisciplinary nature, the field of PdM has been approached from many different angles: this comprehensive survey aims at providing an up-to-date overview focused on all the learning-based industrial PdM strategies, discussing weaknesses and strengths. The survey is based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodological flow, allowing a systematic and complete review of the literature. In particular, firstly, we explore the main learning models used for PdM, mainly Convolutional Neural Networks (ConvNets), Autoencoders (AEs), Generative Adversarial Networks (GANs), and Transformers, also giving an overview of the newest models such as diffusion models and foundation models. Then, we discuss the main learning paradigms applied to PdM, i.e., supervised, unsupervised, ensemble, transfer, federated, and reinforcement learning. Furthermore, this work discusses the pipeline of the data-driven PdM and its benefits, practical applications, datasets, and benchmarks. In addition, the evaluation metrics for each PdM stage and the state-of-the-art hardware devices used are discussed. Finally, the challenges and future work are presented.

Containerization and orchestration have become two key requirements in software development best practices. Containerization allows for better resource utilization, platform-independent development, and secure deployment of software. Orchestration automates the deployment, networking, scaling, and availability of containerized workloads and services. While containerization is increasingly being adopted in the robotic community, the use of task orchestration platforms (e.g., Kubernetes) is still an open challenge. The biggest limitation is due to the fact that state-of-the-art orchestrators do not support real-time (RT) containers, while advanced robotic software often consists of a mix of heterogeneous tasks (i.e., ROS nodes) with different levels of temporal constraints (i.e., mixed-criticality systems). This work addresses this challenge by presenting RT-Kube, a platform that extends the de-facto reference standard for container orchestration, Kubernetes, to schedule tasks with mixed-criticality requirements. It implements monitoring of tasks and detects missed deadlines for those with RT constraints. It selects low-priority tasks to be migrated at runtime to different units of the computing cluster to free resources and recover from temporal violations. We present quantitative experimental results on the software implementing the mission of a Robotnik RB-Kairos mobile robot to demonstrate the effectiveness of the proposed approach. The source code is publicly available on GitHub.

Human-centered robotic applications are becoming pervasive in the context of robotics and smart manufacturing, and such a pervasiveness is even more expected with the shift to Industry 5.0. The always increasing level of autonomy of modern robotic platforms requires the integration of software applications from different domains to implement artificial intelligence, cognition, and human-robot/robot-robot interaction. Developing and (re)configuring such a multi-domain software to meet functional constraints is a challenging task. Even more challenging is customizing the software to satisfy non-functional requirements such as real-time, reliability, and energy efficiency. In this context, the concept of Edge-Cloud continuum is gaining consensus as a solution to address functional and non-functional constraints in a seamless way. Containerization and orchestration are becoming a standard practice, as they allow for better information flow among different network levels as well as increased modularity in the use of multi-domain software components. Nevertheless, the adoption of such a practice along the design flow, from simulation to the deployment of complex robotic applications by addressing the de facto development standards (e.g., ROS - Robotic Operating System) is still an open problem. We present a design methodology based on Docker and Kubernetes that enables containerization and orchestration of ROS-based robotic SW applications for heterogeneous and hierarchical HW architectures. The methodology aims at (i) integrating and verifying multi-domain components since early in the design flow, (ii) mapping software tasks to containers to minimize the performance and memory footprint overhead, (iii) clustering containers to efficiently distribute load across the edge-cloud architecture by minimizing resource utilization, and (iv) enabling multi-domain verification of functional and non-functional constraints before deployment. The article presents the results obtained with a real case of study, in which the design methodology has been applied to program the mission of a Robotnik RB-Kairos mobile robot in an industrial agile production chain. We have obtained reduced load on the robot’s HW with minimal performance and network overhead, thanks to the optimized distributed system.

Analog components are fundamental blocks of smart systems, as they allow a tight interaction with the environment, in terms of both sensing/actuation and communication. This impacts on the design of the overall system, and mainly on the validation phase, that thus requires the joint simulation of digital and analog aspects. In this scenario, this paper proposes the automatic conversion of analog models to C++-based languages, to remove the overhead of co-simulation with traditional virtual platform tools. The proposed methodology allows to convert a given analog description to either: 1) a fully equivalent description or 2) an abstract representation for faster simulation which models only the aspects of interest. Effectiveness and correctness have been proved on a number of case studies that highlight the effectiveness and potentiality of the proposed methodology.