Towards Natural Multi-DoF Prosthetic Control with Distributed Ultrasound

Powered hand prostheses are essential to support limb-difference users living independently. However, prosthesis abandonment rates are high due to a mismatch between the user's needs and the performance of current surface electromyography-powered prostheses. Ultrasound has been proposed as an alternative modality of interfacing, but current work has mainly investigated offline scenarios or virtual benchtop validations. To advance in this area, we designed a custom socket and prosthesis solution with 32 distributed single-element transducers to record line-scanning ultrasound across the stump of one participant. A machine learning model was trained, allowing the participant to voluntarily control simultaneous wrist rotation and hand aperture. Offline cross-validation showed a high performance with an r-square of 0.69 ± 0.26. An Online experiment demonstrated that the participant was able to complete the virtual target achievement control task, the box and blocks test, and the clothespin relocation test in an accurate way. This marks the first functional demonstration of A-mode ultrasound-based simultaneous and proportional control, demonstrating its potential for natural and flexible control of prosthetic hands.