Typically electromyography (EMG) is used to control electro-mechanical prostheses. Below is a demo of some EMG processing software I wrote for controlling a prosthetic.
EMG is fairly robust at identifying large muscle contractions and for a lower arm prosthetic a forearm's contractions and extensions can be reliably used for control. However, this really only gives the user 2 degrees of freedom. Here we explore ultrasound as alternative input and see greatly improved results in terms of both precision and continuous-independent finger control. Some initial pre-processing is performed on the images before training a CNN to predict finger positions.
The resulting model was used to control finger positions of a robotic hand. The ultrasound sensor is mounted to a wrist brace (to avoid movement) targeting the forearm muscles.