Neuromuscular interfacing: a novel approach to EMG-driven multiple DOF physiological models.

This paper presents a novel approach that involves first identifying and verifying the available superficial muscles that can be recorded by surface electromyography (EMG) signals, and then developing a musculoskeletal model based on these findings, which have specifically independent DOFs for movement. Such independently controlled multiple DOF EMG-driven models have not been previously developed and a two DOF model for the masticatory system was achieved by implementing independent antagonist muscle combinations for vertical and lateral movements of the jaw. The model has six channels of EMG signals from the bilateral temporalis, masseter and digastric muscles to predict the motion of the mandible. This can be used in a neuromuscular interface to manipulate a jaw exoskeleton for rehabilitation. For a range of different complexities of jaw movements, the presented model is able to consistently identify movements with 0.28 - 0.46 average normalized RMSE. The results demonstrate the feasibility of the approach at determining complex multiple DOF movements and its applicability to any joint system.

Neuromuscular interfacing: establishing an EMG-driven model for the human elbow joint.

Assistive devices aim to mitigate the effects of physical disability by aiding users to move their limbs or by rehabilitating through therapy. These devices are commonly embodied by robotic or exoskeletal systems that are still in development and use the electromyographic (EMG) signal to determine user intent. Not much focus has been placed on developing a neuromuscular interface (NI) that solely relies on the EMG signal, and does not require modifications to the end user's state to enhance the signal (such as adding weights). This paper presents the development of a flexible, physiological model for the elbow joint that is leading toward the implementation of an NI, which predicts joint motion from EMG signals for both able-bodied and less-abled users. The approach uses musculotendon models to determine muscle contraction forces, a proposed musculoskeletal model to determine total joint torque, and a kinematic model to determine joint rotational kinematics. After a sensitivity analysis and tuning using genetic algorithms, subject trials yielded an average root-mean-square error of 6.53° and 22.4° for a single cycle and random cycles of movement of the elbow joint, respectively. This helps us to validate the elbow model and paves the way toward the development of an NI.