Work-related musculoskeletal disorders are globally one of the leading causes of work-related injuries. They significantly impact worker health and business costs. Work task ergonomic risk indices have been developed that use observational assessments to identify potential injuries, and allow safety managers to promptly intervene to mitigate the risks. However, these assessments are very subjective and difficult to perform in real time. This work provides a technique that can digitalize this process by developing an online algorithm to calculate the NIOSH index and provide additional data for ergonomic risk assessment. The method is based on the use of inertial sensors, which are easily found commercially and can be integrated into the industrial environment without any other sensing technology. This preliminary study demonstrates the effectiveness of the first version of the Online Lifting Index (On-LI) algorithm on a common industrial logistic task. The effectiveness is compared to the standard ergonomic assessment method. The results report an average error of 3.6% compared to the NIOSH parameters used to calculate the ergonomic risk and a relative error of the Lifting Index of 2.8% when compared to observational methods.

The assessment of realistic work tasks is a critical aspect of introducing exoskeletons to work environments. However, as the experimental task’s complexity increases, the analysis of muscle activity becomes increasingly challenging. Thus, it is essential to use metrics that adequately represent the physical human–exoskeleton interaction (pHEI). Muscle activity analysis is usually reduced to a comparison of point values (average or maximum muscle contraction), neglecting the signals’ trend. Metrics based on single values, however, lack information about the dynamism of the task and its duration. Their meaning can be uncertain, especially when analyzing complex movements or temporally extended activities, and it is reduced to an overall assessment of the interaction on the whole task. This work proposes a method based on integrated EMGs (iEMGs) to evaluate the pHEI by considering task dynamism, temporal duration, and the neural energy associated with muscle activity. The resulting signal highlights the task phases in which the exoskeleton reduces or increases the effort required to accomplish the task, allowing the calculation of specific indices that quantify the energy exchange in terms of assistance (AII), resistance (RII), and overall interaction (OII). The method provides an analysis tool that enables developers and controller designers to receive insights into the exoskeleton performances and the quality of the user-robot interaction. The application of this method is provided for passive and two active back support exoskeletons: the Laevo, XoTrunk, and StreamEXO.

Inspections of industrial and civil infrastructures are necessary to prevent damages and loss of human life. Although robotic inspection is gaining momentum, most of the operations are still performed by human workers. The main limiting factors of inspection robots are the lack of versatility as well as the low reliability of these devices, since they need to operate in a non-structured environment. In this work, a novel Hybrid Platform for inspection in industrial contexts is proposed, focusing on the design and testing of the Crawler Unit. The goal is to solve versatility related issues exploiting modularity and self-reconfigurability. The Hybrid Platform consists of three main systems: a mobile Main Base and two Crawler Units. Each would operate independently, accomplishing specific tasks. Docking interfaces, on each device, allow the systems to reconfigure into different robots. The Crawler Unit operates in constrained environments and narrow spaces. The Main Base patrols wide areas and deploys the Crawler Units near the inspection site. For dealing with challenging conditions, the two Crawler Units can dock together, reconfiguring into a snake-like robot. Additionally, once docked to the Main Base, the two Crawlers can operate also as robotic arms, providing manipulation abilities to the platform. The first version of the Crawler Unit exhibited an interesting performance over flat and uneven terrains. To extend the mobility of this robot, a second version was developed, introducing some innovations in the system design. These innovations provided the Crawler Unit with advanced mobility in the vertical plane, thus allowing the robot to deal with more complex scenarios such as crossing gaps, overcoming obstacles and lifting the modules.

Human Activity Recognition (HAR) is increasingly used to enhance exoskeleton control, collaborative robotics, and ergonomic assessments by providing high-level motion insights for industrial applications. However, deploying HAR on portable edge devices with limited computational resources presents challenges, highlighting the need for lightweight algorithms tailored to industry. This study proposes and compares low-powered lightweight HAR models capable of real-time operation on portable wearable systems. The models exploit sensor fusion from a full-body wearable system, the Smart Suit, designed for construction workers, with the goal of enabling accurate real-time activity recognition while preserving the Natural Human Motion Frequency (NHMF) of 5Hz. The impact of window size selection and the inclusion of transitional activities was analyzed to evaluate multiple Machine Learning (ML) (SVM, NN, Ensemble) and Deep Learning (DL) (CRNN, BiCRNN, BiLSTM, CNNTransformer) models combined with signal optimization techniques. Performance was assessed in terms of accuracy, F1-score, training time, inference time, FLOPs, energy consumption, and model size. A dedicated industrial dataset was iteratively constructed using a novel Elementary–Transition–Elementary (ETE) recording technique, which simplified the recording and labeling of complex tasks and supports dataset scalability. Model generalization was evaluated using a Leave-One-Subject-Out protocol, demonstrating that selected models achieved a maximum accuracy of 92.7% under realistic deployment constraints, enabling up to 10 classifications per second.