Learning-Based Approach to Assist Occupational Workers Using Soft Pneumatic Wrist Exoskeleton

Musculoskeletal disorders are a common concern among occupational and healthcare workers, yet many cases can be mitigated with the use of exoskeleton-based assistive technologies. Soft robotics has significantly advanced exoskeleton design by enhancing safety, ergonomics, and adaptability. However, the nonlinear properties and hysteresis of soft robotic devices pose challenges in modeling and control. To overcome these limitations, learning-based approaches have gained traction as effective solutions. This study introduces two learning-based models for a pneumatically actuated soft wrist exoskeleton aimed at supporting occupational workers and reducing the risk of musculoskeletal disorders. The exoskeleton incorporates a pneumatic soft actuator that enables wrist flexion and extension. The first model utilizes machine learning (ML) techniques to establish a correlation between air pressure, wrist bending angle, and the percentage of hand weight supported by the exoskeleton. This correlation allows for the estimation of the required input air pressure for the actuator to achieve a given bending angle while assisting a known amount of weight. Several ML models are evaluated and compared based on their accuracy and computational efficiency. The second model focuses on predicting the percentage of valve opening to regulate the airflow rate, ensuring that the wrist joint moves at a desired velocity while assisting a specified weight to reach a desired angle.