Artificial Neural Network that Modifies Muscle Activity in Sit-to-Stand Motion Using Sensory Input

Abstract

Sit-to-stand motion is an important daily activity, and it is important to study the mechanism of the motion to improve the ability when it becomes weak. To study the mechanism, we hypothesized that muscle synergy generates muscle activity as a feedforward signal, which is modified by sensory input. This study focuses on determining the sensory input primarily used for modifying sit-to-stand motion. To obtain this, we built artificial neural network models that generate muscle activities based on sensory input and feedforward signals and analyzed the effect of each input on the output. The models were built for each motion phase. The input was information from vestibular and somatosensory input and averaged muscle synergy as feedforward signals, and the output was muscle synergy. As a result, it was revealed that humans may primarily use hip angle to bend forward, ankle and vertical foot reaction force to hip rise, ankle, knee, and lumber angles and vertical foot reaction force to extend body, and lumber angle to stabilize. This indicates the type of sensory input used to control each muscle synergy in each motion phase. The information should be used to modify the sit-to-stand motion in environmental conditions where the motion is performed.