Leader: Fanny Ficuciello (UNINA); Other collaborator(s):
The goal is to identify personalised outcomes of robotic technology in rehabilitation of older multimorbid patients to inform spoke 8 activities in developing and testing robotic systems aimed at improving the quality and ability of movement of the elderly and frail people in the execution of daily tasks. Ad hoc control algorithms will be developed for elderly and frail subjects and testing protocols will be produced. The control strategies will be adapted according to the task, both for rehabilitation and for daily assistance, and will take into account the ascertained physical conditions of the patient as well as the physiological signals in real-time.
Brief description of the activities and of the intermediate results: In collaboration with Dr. Raffaele Dubbioso of the Department of Neuroscience of the Federico II University of Naples we draft an experimental protocol to be submitted to the Ethics Committee. Subject that will be involved are Amyotrophic Lateral Sclerosis (ALS) patients (of the Flail Arm and Flail Leg type), whom the doctor involved in the study is in charge of, and whose average age exceeds 65 years, associated with a group of healthy subjects as a reference. The analysis of the performance of these subjects will serve as a guideline to define the best outcomes of interest once robotic rehabilitation systems are applied.
Considerations from the analysis of the performance of a healthy subject using an end-effector rehabilitation system were integrated into the process of drafting and evaluating the setup and procedure to be followed in the experimental protocol. In particular, performance was assessed by taking into account the subject's muscle activations from information recorded by EMG sensors. In addition, specific algorithms were developed to control and assist in the completion of the tasks performed, taking into account the muscular-skeletal model of the upper limb and the subject's activation capabilities. The procedures to be followed in the experimental protocol agreed upon with Dr. Dubbioso were defined and drafted in full. In detail, the EMG sensors to be used, the muscles considered to be of greatest interest in assessing the motor skills of the test subjects, and the filtering techniques to be used on the raw data were chosen. In addition, an IMU sensor was also identified to be included within the setup, so as to derive information regarding the motion of the test subject's limb in terms of position, velocity, and acceleration. A specific video capture system was also included, with the purpose of mapping, through specific markers, the positions in space of all the joints of the limb and making a more complete assessment of the motion. In addition to the choice of setup, the metrics of greatest interest and relevance that will be extracted from these data to go on to numerically categorize the residual abilities of the patients under study were then also defined.
Drafting of the paper on HL-EU-Q6 validation for older people.
Main policy, industrial and scientific implications: These activities will be useful in identifying suitable robotic technology for specific diseases of the older people.
The previously drafted experimental protocol underwent corrections requested by the Ethics Committee. After the changes were implemented, the new version of the protocol was evaluated and officially approved in May. This allowed the experimental activities to begin in collaboration with the doctors who will follow the subjects involved in the study. The activities involved the collection of data from healthy subjects.
In parallel, algorithms were developed to analyze and extract metrics of interest from the collected data. In detail a vision algorithm was developed to analyze, segment and track markers used for mapping the movement of human joints while performing the task of interest. Segmentation of task sub-phases was also initiated using algorithms based on data extracted from the IMU sensors to allow for more detailed analysis. In addition, the exercises required to correctly record the Maximum Voluntary Contraction (MVC), a key parameter for meaningful analysis of EMG signals, were refined.
A literature review phase was initiated to identify the most effective control strategies to support the rehabilitation of the subjects. The integration of biosignals into the exoskeleton control loop was also investigated. With regard to passive exoskeletons, a study was initiated to identify the optimal configuration of auxiliary sensors to assess their effectiveness in subjects with functional limitations.
Additionally a new system, for evaluating the walking performance of the subjects involved in the experimental protocol developed, was integrated into the data acquisition setup. The system consists of two sensorised platforms for the measurement of the force applied to the ground by the subject under examination during step execution.
Finally, a bibliographic investigation phase has been initiated in order to define innovative quality indices, that take into account various bio-signals and parameters recorded during acquisitions. The final aim of this research is to extract a hyper-parameter indicative of the physiological state of the subject undergoing the specific physiotherapy assessment.
Data acquisition from healthy subjects has been ongoing, while initial data from patients with Amyotrophic lateral sclerosis (ALS), specifically in the early stages of the disease and presenting a Flail Arm phenotype, were also gathered. A passive upper limb exoskeleton was applied to the ALS subjects as part of this data collection process. The algorithms designed for data analysis were developed based on bibliographic research, and quality indices were derived to evaluate the data.
As the dataset grew with the inclusion of new ALS patients, the statistical analysis was updated, allowing for a more comprehensive comparison between healthy subjects and ALS patients. This expanded dataset enabled a more detailed evaluation of the differences and potential trends between the two groups.
Additionally, new algorithms were introduced to improve data analysis processes, and ongoing statistical analysis allowed us to refine the results. Preliminary results from these analyses have contributed to the preparation of a scientific paper.
Moreover, evaluations were carried out regarding the integration of a motor into one of the active exoskeleton structures within the experimental framework. These evaluations focused on potential design modifications aimed at enhancing the exoskeleton’s functionality. This work is being integrated into the broader effort to develop both passive and active exoskeleton systems for improved support in ALS patients.
In parallel, a literature search on the study of muscle synergies has begun in collaboration with a neurologist. The aim is to classify muscle synergies according to the stage of the disease and the muscles most involved. This approach is based on the idea that, as the disease progresses, patients tend to compensate for the loss of function in specific muscles through new neuromuscular activation strategies. Studying these residual synergies could provide crucial information on the compensatory mechanisms that are activated in response to motor deterioration.
During the period October–December 2024, advancements were made in both data collection and analysis to deepen the understanding of motor impairment and muscle fatigue in ALS patients and healthy controls. Data collection efforts focused on ALS patients with the Flail Arm (FA) phenotype, including reassessments of previously monitored patients to track the progression of muscle weakness. Concurrently, data collection from healthy subjects provided baseline information to support comparative analyses.
The experimental protocol was refined to enhance the evaluation of muscle fatigue. Based on preliminary observations, adjustments were made to the structure and execution of exercises to ensure tasks more effectively captured fatigue-related changes in muscle activity. These modifications aimed to provide more accurate and detailed fatigue metrics, which are critical for understanding motor impairment progression in ALS patients.
In parallel, data analysis was conducted on surface electromyography (sEMG) signals, focusing on extracting muscle synergies. Various methodologies were explored to identify the optimal number of synergies, a key parameter for assessing motor control capacity and its degradation in pathological conditions.
Additional physiological signals were integrated to enrich the experimental framework further, including electrocardiographic (ECG) data and galvanic skin response (GSR). These enhancements aim to provide a comprehensive perspective on muscle fatigue by correlating it with broader physiological responses.
Data collection and analysis activities continued during the reporting period, involving healthy subjects and ALS patients. The acquisition focused on surface electromyography (sEMG), electrocardiographic (ECG), and galvanic skin response (GSR) signals. These data are currently under analysis to support comparative assessments of motor function and muscle fatigue across groups. Ongoing analysis of sEMG signals included the extraction of muscle synergies. Multiple methods were evaluated to determine the optimal number of synergies, a parameter relevant to assessing motor control and its deterioration in neurodegenerative conditions.
Additionally in these months, significant efforts were dedicated to the development of a modified version of the Upper Limb EDU Exo. The entire system underwent a complete mechanical and electronic redesign to improve functionality, control, and reliability. All components were re-engineered in SolidWorks, focusing on weight reduction, subsystem integration, and mechanical optimization. Particular attention was given to integrating the RMD motor into the shoulder joint, ensuring a stable and secure connection while minimizing structural modifications and maintaining a lightweight design. Key components were prototyped using 3D printing with Onyx material, selected for its high strength-to-weight ratio. Initial mechanical tests indicated the need for minor adjustments, which were promptly implemented to improve overall performance. Communication between the hardware and the control architecture is being established using Arduino and ROS 2, ensuring compatibility with the existing research platform. IMU sensors were installed, and preliminary calibration procedures were completed, enabling motion tracking and orientation feedback. Current efforts are focused on finalizing the hardware setup and ensuring reliable data acquisition from all integrated sensors.
This reporting period also included the design and preliminary prototyping of a hand exoskeleton (ExoGlove). The first phase focused on developing a personalized design through a 3D scan of the user’s hand, which served as the anatomical basis for the CAD model. The model was created and iteratively refined using Solid Edge to align with ergonomic and functional requirements. To further improve the mechanical design and actuation layout, OpenSim—an open-source biomechanical simulation tool—was used to model the musculoskeletal dynamics of the human hand. This simulation provided key insights into muscle-tendon interactions, joint movements, and force transmission, which informed structural refinements aimed at improving both comfort and functional efficiency. This biomechanics-informed design process is expected to significantly improve the device’s performance in grasping tasks, particularly in rehabilitation and assistive scenarios.
During the past three months, the research on the Upper Limb Edu-Exo project has advanced significantly, particularly in terms of mechanical refinements, control strategies, and sensor integration. A series of mechanical improvements were introduced to optimize the ergonomics and structural balance of the exoskeleton. In particular, the motor shaft length was reduced, resulting in a better weight distribution and improved comfort for the user during prolonged wear. Additionally, a new shoulder motor with higher torque capabilities was selected and integrated to enhance the system’s ability to execute dynamic and load-demanding tasks. A major research project focused on the development of an adaptive impedance control framework based on Reinforcement Learning. In particular, the ongoing research is focusing on the implementation of the i-MOGIC (Integrated Motion Generator and Impedance Controller) on the exoskeleton. This approach replaces the traditional dual-loop architecture with a unified formulation that combines motion generation and impedance control within a single dynamical system. The core idea behind i-MOGIC is to represent the exoskeleton’s behavior as a combination of local virtual springs and dampers, distributed throughout the state space. The stiffness and damping properties of these local MSD systems are not fixed, but are instead learned through Reinforcement Learning, allowing the system to adapt to the user’s needs automatically. The learning process is guided by a Cross-Entropy Method (CEM)-based optimization, which ensures that all impedance parameters remain stable and physically consistent throughout training. This structure enables the exoskeleton to regulate both movement and interaction forces in a compliant and adaptive way, with formal guarantees of stability and robustness to disturbances.
In parallel with the development of the Upper Limb EDU Exoskeleton, progress has also been made on the ExoGlove project, a soft hand exoskeleton designed to support fine motor functions. The current phase focuses on enhancing its adaptability, actuation system, and control strategies. Building upon the initial biomechanics-informed design of the ExoGlove, recent efforts have focused on advancing the project’s engineering structure and control architecture to enable broader adaptability and improved functionality. A key development area is the transition from a user-specific CAD model to a parametric design framework. Work is currently underway to define a set of anthropometric parameters that will serve as driving variables within Solid Edge. This parametric modeling approach aims to create a scalable and customizable exoskeleton design, significantly reducing adaptation time for different users while ensuring anatomical consistency and comfort. In parallel, an extensive study is being conducted to identify the most suitable actuation system for the ExoGlove. The analysis is primarily focused on two candidate architectures:
The final choice will balance mechanical simplicity, safety, and controllability, based on both simulation and prototype-based testing. Simultaneously, efforts are directed toward enhancing the human-machine interface by investigating which biological signals are most suitable for decoding user intent. Current experiments are centered on electromyography (EMG) to interpret voluntary
muscle activations and mechanomyography (MMG) as a potentially more robust and complementary signal, particularly in noisy or variable EMG acquisition contexts. The convergence of these three development axes—a parametric CAD model, optimized actuation, and bio-signal-based control—is expected to result in a highly versatile, user-centered hand exoskeleton capable of supporting various rehabilitation and assistive scenarios.
During the past months, research activities on the Upper Limb Edu-Exo project have primarily focused on advancing the dynamic characterization and observer-based friction estimation of the exoskeleton’s actuated joints. This phase aimed to enhance the transparency and control accuracy of the system, ensuring smoother human–robot interaction and improving low-speed behaviour during assistive motion. One of the major developments during this period was the implementation and experimental validation of a momentum-based friction estimation framework. The goal was to identify and compensate for nonlinear friction effects that limit transparency and smoothness, particularly near zero velocity where stick–slip phenomena are prominent. Two estimation schemes were implemented and compared under identical conditions:
• A First-Order Momentum Observer (FO) used as a linear baseline.
• A Second-Order Sliding-Mode Momentum Observer (SOSML) designed for higher robustness and faster convergence.
Both observers were integrated into the two-DoF Edu-Exo platform, modeled through Simscape Multibody to preserve realistic joint dynamics and reaction forces. The observers reconstruct joint friction in real time using only proprioceptive signals (joint positions, velocities, and motor currents), avoiding the need for additional sensors. To validate the approach, three excitation trajectories were designed to cover different operating regimes:
1. Speed Ladder (T1) – sequences of constant-velocity plateaus to estimate viscous and Coulomb friction components;
2. Symmetric Reversal (T2) – frequent direction changes to capture low-speed Stribeck behaviour;
3. Slow Ladder (T3) – extended low-speed plateaus for bias and steady-state validation.
Experimental data confirmed that both observers provided accurate friction reconstruction on steady velocity plateaus. However, the SOSML observer exhibited tighter phase-plane loops, faster convergence after reversals, and superior noise robustness, particularly in the low-speed Stribeck region.
The consolidated friction parameters obtained from SOSML accurately reproduce the nonlinear S-shaped friction law across all trajectories and will serve as the reference model for future compensation strategies.
By reducing estimation lag and improving noise rejection without added hardware, the system now achieves a higher level of transparency and control precision—critical for human-in-the-loop assistance and rehabilitation tasks.
In parallel with the activities carried out on the Edu-Exo system, progress has also been achieved on the ExoGlove project, particularly in the areas of parametric modeling and automation of the design workflow. A further advancement in this development phase concerned the creation of a Python-based processing pipeline dedicated to acquiring and analyzing anatomical measurements obtained from 3D scans of the physical hand cast. The script automatically extracts the key geometric and morphological parameters and integrates them directly into the parametric CAD model of the device.
Through this automated workflow, a first fully parameterized prototype of the ExoGlove system was generated in SolidWorks, allowing the overall dimensions and structural features to be rapidly adjusted by modifying only a limited number of input variables.
This methodology significantly reduces the adaptation and customization time required for different users, while ensuring higher dimensional accuracy and functional consistency between the digital model and the real hand. Moreover, the automation of the parameterization process represents an important step toward scalable and semi-automated production, enabling direct configuration of the device from real biometric data. The integration of parametric modeling, automated measurement acquisition, and adaptive actuation principles therefore establishes the foundation for a flexible, customizable, and user-centered system, fully aligned with the project’s overarching goals of versatility and personalization.
During the last months of the project, the focus was mainly on the ExoGlove project. Building upon the automated parameterization workflow developed in the previous quarter, research activities shifted toward architectural optimization, structural validation, and functional prototyping of the wearable device.
Bio-inspired Actuation System: A central development of this final phase was the design and implementation of a bio-inspired actuation system. To overcome the bulk and weight constraints of classical systems, a Multiport-Output Twisted String Actuator (MO-TSA) configuration was adopted. This underactuated solution utilizes a single torque source to drive four fingers through two output branches (Index-Middle and Ring-Little), mimicking the human hand’s flexor muscle architecture. This architectural choice enables adaptive behavior under differential loading, significantly reducing mechanical complexity while improving portability for rehabilitation tasks, following the principles of polymer-based wearable robotics.
Numerical Optimization and Sizing: To ensure optimal performance, a numerical optimization procedure was conducted in MATLAB to define the actuator’s key geometric parameters. The optimization focused on two primary design variables:
The resulting configuration achieved a required displacement (Δx) of over 70mm within 29.18 turns, providing an actuation time of 15.9 s and an available force per finger of approximately 11N. This meets the design constraints for functional assistance, aligning with state-of-the-art TSA-powered biomimetic gloves.
Structural Integrity and Material Selection: In parallel, the structural integrity of the device was assessed through extensive Finite Element Method (FEM) analysis. This phase focused on selecting materials to balance rigidity, flexibility, and user comfort, a critical factor for stable wearing in grasping assistance. Multiple iterations were performed to evaluate 3D-printed polymers, including:
• Standard PLA and Carbon Fiber-reinforced PLA (PLA CF) for rigid structural components.
• TPU (85A and 95A) for flexible joints and interface elements.
The FEM results led to the refinement of Finger Design 2.0, which optimizes force distribution across the MCP, PIP, and DIP joints. The analysis confirmed that the combination of PLA CF and TPU provides the necessary durability to withstand the 10N peak loads required during grasping tasks.
Experimental Validation: The project concluded with the integration of the actuator case and functional testing. Validation focused on the available workspace and grasping motions (spherical and cylindrical). Workspace estimation was conducted using the Manus VR glove to provide a benchmark against physiological literature. In parallel, biological signal acquisition was investigated to assess the feasibility of user-driven control strategies for the ExoGlove.
Comparison of Biological Signals (EMG vs MMG): A comparative analysis of electromyographic (EMG) and mechanomyographic (MMG) signals was carried out to support the experimental validation phase and to evaluate their suitability for controlling the ExoGlove. The comparison links muscle activation to functional hand movements by relating neural command generation to the resulting mechanical muscle response. Raw EMG signals exhibit low amplitudes (10−2–10−1 mV) and a spectrum dominated by high-frequency components, resulting in a noisy appearance that persists even after band-pass filtering. Although these components convey information related to motor unit recruitment, their spectral overlap with electrical noise and environmental interference complicates reliable feature extraction, particularly in wearable and dynamic conditions. MMG signals, in contrast, present higher amplitudes and slower, more regular oscillations associated with muscle mechanical activity. After filtering, motion-related artifacts are effectively attenuated, while muscle-induced vibrations remain clearly distinguishable, resulting in a signal that is easier to process and more robust for real-time applications. The analysis of time- and frequency-domain features confirms these differences. Amplitude-based features such as Root Mean Square (RMS) and Mean Absolute Value (MAV) show coincident peaks in EMG and MMG during intense contractions, indicating a strong temporal correlation between electrical activation and mechanical response. However, MMG exhibits significantly higher absolute values and reduced sensitivity to noise. Similarly, signal variance (VAR) is more interpretable in MMG, as it directly reflects mechanical instability and tremor, whereas EMG variance is strongly influenced by high-frequency noise. Frequency-domain features further differentiate the two modalities. EMG Mean Frequency (MF) typically ranges between 20 and 40 Hz, while MMG MF is concentrated between 10 and 30 Hz, reflecting slower mechanical oscillations. Variations in Zero Crossing Rate (ZCR) are also more informative in MMG, enabling clearer observation of tremor and mechanical instability, which are less evident in EMG due to electrical interference. Overall, MMG demonstrates reduced susceptibility to electrical noise and electrode-related artifacts, providing a more stable and interpretable signal in dynamic and wearable scenarios such as the ExoGlove.
Implications for ExoGlove Control Strategies: The comparative evaluation of EMG and MMG signals directly informs the control architecture of the ExoGlove. Since the device is designed to deliver adaptive and user-centered assistance during grasping tasks, the choice of biological input signal significantly affects robustness, responsiveness, and usability. An EMG-based control strategy enables early detection of user intent by exploiting the direct link between neural activation and muscle recruitment. While this allows potentially fast and anticipatory control, EMG is highly sensitive to noise, electrode placement, and skin conditions, often requiring complex filtering and adaptive processing to maintain control stability. A MMG-based control approach, instead, relies on the mechanical response of the muscle and offers increased robustness against electrical interference. The strong correlation between MMG amplitude- and variancebased features and contraction intensity makes MMG well suited for proportional and continuous control schemes. Although MMG reflects activation with a slight delay compared to EMG, this limitation is acceptable in assistive grasping tasks, where smoothness and reliability are prioritized. A hybrid EMG–MMG strategy represents a particularly promising solution. In this configuration, EMG can be used to detect movement onset and user intention, while MMG provides a stable estimate of contraction strength and mechanical output. This multimodal approach enhances control robustness and adaptability, aligning well with the underactuated and compliant nature of the MO-TSA architecture adopted in the ExoGlove. In conclusion, the integration of EMG, MMG, or their combination enables a flexible and scalable control framework. The results suggest that MMG-based or hybrid strategies provide the most effective trade-off between responsiveness and robustness for real-world deployment of the ExoGlove.
Final Remarks: These final steps establish a robust foundation for a scalable, semi-automated production workflow, transforming biometric 3D scans into a personalized orthotic device. Through the integration of MO-TSA actuation and parametric design, the system delivers a lightweight, user-centered solution for upper limb rehabilitation. In parallel, the experimental investigation of EMG and MMG signals defines a flexible framework for user-driven control, paving the way for real-time implementation of EMG-, MMG-, or hybrid-based strategies with adaptive assistance, user-specific calibration, and robust performance during prolonged and dynamic use.
Dissemination
Accepted Publications