Traumatic Brachial Plexus Injury (TBPI) causes arm paralysis, impacting daily activities and reintegration into work. This paper presents the results of the first round of a quantitative Delphi study. Aim is to gather insights from five experts groups. The investigated user requirements primarily encompass design aspects like weight, fitting, and adjustability to accommodate individual adaptations. Besides increased functionality a benefit for psychological recovery is presumed.

Introduction

Neurodegenerative diseases such as Ataxia or Parkinson’s can significantly affect daily activities such as eating and drinking due to motor control impairments, including tremors or overshooting arm movements. Wearable as-sistive devices offer a promising solution to overcome these challenges. However, real-time estimation of correc-tive forces requires the detection and prediction of the intended and dysfunctional movement components. The research project aims to develop such a lightweight assistive device capable of actively applying corrective forc-es to only suppress dysfunctional movement enabling the intended movement.

Methods

To achieve this goal, the arm trajectories during activities of daily living need to be predicted and divided into intended and dysfunctional movements with their corresponding forces. Our approach uses inertial sensors and video recordings to detect the intentional and pathological movement in real-time. On the other hand, we use a computationally efficient neuro-musculoskeletal arm model to quantify the human upper limb stiffness respons-es in different postures. Furthermore, we employ reinforcement learning (RL) to simulate realistic arm move-ments, accounting for noise, optimality principles, and various task requirements.

Results

Preliminary results show that we can reliably detect the frequency and amplitude of tremorous movement com-ponents. Our numerical arm model accurately predicts human upper limb impedance characteristics, closely matching experimental data. Furthermore, our initial results indicate that combining RL with a neuro-musculoskeletal arm model can replicate highly stereotypical arm-reaching movements, including straight hand trajectories, bell-shaped tangential velocity profiles, and triphasic muscle activation patterns.

Conclusion

In conclusion, we have shown the feasibility of detecting dysfunctional movement and predicting both voluntary actions and necessary corrective forces using our approach. In our future work, these predictions will be integrat-ed into real-time estimation for assistive devices, with the ongoing development of an intelligent controller pipe-line to integrate all components effectively.

Industrial exoskeletons have recently gained importance as ergonomic interventions for physically demanding work activities. The growing demand for exoskeletons is leading to a need for new knowledge on the effectiveness of these systems. The Exoworkathlon, as a prospective study approach, aims to assess exoskeletons in realistic use cases and evaluate them neutrally.

Different test scenarios (Parcours) were developed with experts from the corresponding industries for the stand-ardized and realistic evaluation of industrial exoskeletons. The Parcours represent a real work-scenario abstract-ly. Till now, Parcours from Logistics Automotive , Construction Work, and Welding have been developed. The subjects are young experts from the corresponding area. A working time of 1 hr is chosen to ensure that the Par-cours are realistic and feasible. Each participant runs the corresponding Parcour twice - 1 hr with and 1 hr with-out an exoskeleton, with a break of at least 2 hrs to recover. The order of exoskeleton conditions is randomized, and the exoskeletons are randomly assigned to the subjects. Recommended assessment parameters are muscle activity, metabolics, subjective user feedback and quality aspects.

So far, 125 subjects have been tested on six Parcours. The effort of the task during the activity is reduced both Parcour-specifically and across all Parcours. This is also shown by the reduction in muscle activity (n=21). The usability (SUS scale) is rated as “good” across all exoskeletons. The welding quality (n=41) can be significantly improved, and a trend towards quality improvement can also be seen in the drywall Parcour (n=9).

The 125 test subjects' data show potential relief and quality improvement in certain use cases. In future musculo-skeletal modelling can be used for detailed biomechanical analysis and to detect possible side effects.

Exosuits represent a promising technology for upper limb assistance, offering innovative solutions to enhance motor function in individuals with neuromuscular impairments. Nevertheless, challenges persist in seamlessly integrating exosuits with human motion and navigating interactions with the environment. The integration of computer vision, especially object recognition algorithms, offers a solution by enhancing exosuits' environmental awareness and adaptability in adjusting assistance levels. In this study, we developed an algorithm employing computer vision to fine-tune the assistance level of an elbow exosuit, effectively reducing joint stress during object interactions. Validation of the algorithm demonstrated a significant decrease in muscle strain during dynamic activities, consistently correlating with object mass. With an average object recognition accuracy of 93.5%, this work sets the stage for integrating the vision algorithm, thus improving user-exosuit interactions, including adaptability to environmental dynamics.