Christian Di Natali

To promote the adoption of exoskeletons in industries, they must be effective in providing physical assistance; however, other factors, such as familiarity with their use and comfort, contribute to device acceptance and workers’ approval. During on-site testing, the use of common assessment tools (e.g., EMGs, MoCap systems, etc.) and the pool of available subjects are usually limited. Thus, evaluating the reliability of qualitative results (e.g., from questionnaires) is critical. This paper investigates the impact of introducing a back-support exoskeleton to five construction workers in operative scenarios. Key factors, including perceived physical exertion, the task workload, and overall exoskeleton acceptance, were evaluated. The results were compared before and after the workers were instructed on the proposed familiarization approach. This analysis aims to investigate the potential influence of novelty bias by assessing the consistency of users’ feedback, which shows increased agreement among workers after familiarization. The relative dispersion of responses is assessed using the Coefficient of Variation (CoV) index. The evidence suggests a potential bias and overestimation in the results obtained during the first test day (before familiarization). After familiarization, the results indicate incremental reliability for the selected questionnaires. Physical exertion perceived in the trunk (before and after familiarization) is further reduced by 0.5 out of 5, while acceptance of the device increases from 5.7 to 6.3 out of 7. The proposed method can support the adoption of industrial exoskeletons by providing a tool for assessing the reliability of qualitative data.

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.

Versatile exoskeletons are active occupational exoskeletons that can recognize and consequently support specific user activities. This feature makes them suitable for assisting complex activities such as manual material handling (MMH), which involves sequences of static, dynamic, and cyclical tasks that can be combined. Active exoskeleton control requires a level of activity recognition that extends beyond the simple recognition of basic activities, such as standing, bending, and locomotion. When the atomic actions associated with MMH are identified, developing advanced and task-specific assistive strategies becomes possible. This work proposes a hierarchical classification algorithm, designed in three layers, to classify the tasks performed and their atomic actions. The algorithm blends information from EMGs and IMUs, using multi-modal logistic regressors and rule-based classifiers. The algorithm classifies locomotion by distinguishing between hip flexion and extension phases (90.8% accuracy), bending by distinguishing between stoop and squat techniques (91.1% accuracy), and the presence of payload hold on the subject's hands (95.6% accuracy). In addition, the algorithm can distinguish between walking and carrying, lifting and unloaded bending, etc. The algorithm is user-agnostic and was trained, validated, and tested on eighteen subjects who handled different loads and performed different task sequences.

Anchor point placement is a crucial yet often overlooked aspect of exosuit design since it determines how forces interact with the human body. This work analyzes the impact of different anchor point positions on gait kinematics, muscular activation and energetic consumption. A total of six experiments were conducted with 11 subjects wearing the XoSoft exosuit, which assists hip flexion in five configurations. Subjects were instrumented with an IMU-based motion tracking system, EMG sensors, and a mask to measure metabolic consumption. The results show that positioning the knee anchor point on the posterior side while keeping the hip anchor on the anterior part can reduce muscle activation in the hip flexors by up to 10.21% and metabolic expenditure by up to 18.45%. Even if the only assisted joint was the hip, all the configurations introduced changes also in the knee and ankle kinematics. Overall, no single configuration was optimal across all subjects, suggesting that a personalized approach is necessary to transmit the assistance forces optimally. These findings emphasize that anchor point position does indeed have a significant impact on exoskeleton effectiveness and efficiency. However, these optimal positions are subject-specific to the exosuit design, and there is a strong need for future work to tailor musculoskeletal models to individual characteristics and validate these results in clinical populations.

Physical fatigue is a major cause of work-related accidents and musculoskeletal injuries in the construction industry, and additional means are needed for their identification and management to prevent long-term consequences. Based on recent scientific literature, fatigue can be detected with wearable inertial measurement units (IMUs). However, IMUs for detecting fatigue have been so far tested mainly in the laboratory; therefore, a research gap exists in application of IMU sensors for detecting fatigue in real-life work settings. The aim of this paper is to bring the fatigue trend detection with IMUs closer to real-life context by using wearable IMU sensor data from an actual construction site measuring actual workers with simulated work tasks. The paper also presents advancements in fatigue trend detection with frequency domain investigations to gain access to more detailed fatigue relevant features. Machine-learning methods are used to predict fatigue trends based on IMU data, resulting in fatigue trend detection accuracy that advances the state of the art. More knowledge is also unearthed about relevant sensor locations and features.