Muscle synergies for online simultaneous control of two DOFs of a robotic arm

Myoelectric control can significantly improve human–robot interaction due to the ability to non-invasively measure human motion intent. Intensive research has worked on the attempt of providing the user with simultaneous and intuitive control of multiple Degrees of Freedom (DOFs). Here, we propose a method based on the theory of muscle synergies, which assumes that movements are executed by translating low-dimensional task-level neural commands into high-dimensional muscle activations. Our approach models each DOF as driven by 2 activation signals and it estimates neural control information from 8 muscles of the upper-limb both during movement of the shoulder (DOF 1) and elbow (DOF 2) joints in the horizontal plane (Dynamic Condition), and during isometric contractions (Isometric Condition). The algorithm requires a calibration phase (Day 1), where the subject is asked to activate the 2 DOFs separately. A DOF-wise nonnegative matrix factorization (NMF) is then applied to extract a subject-specific synergy matrix (S). In the control phase, the subject directly uses the activation signals (extracted from the online EMG and the S matrix by solving a nonnegative least-squares constraints problem) for online torque-based control of 2 DOFs of a robotic arm. Low-latency is achieved by introducing a nonlinear filtering of the EMG based on Bayesian estimation. To test the robustness and repeatability of the algorithm over days, the control phase is tested also the following day (Day 2) using the same S extracted from the calibration phase on Day 1. To assess the effectiveness of our approach compared with the traditional single-muscle method, where 2 independent muscles are used for each DOF separately, we perform offline simulations where each DOF of the robotic arm is driven by 2 EMG signals that correspond to the muscles most involved in each of the 2 directions of articulation of the joint. For the Dynamic Condition, the performance of the control algorithm is assessed by computing the Root-mean-square Error (RMSE) and the Pearson’s Correlation coefficient (R) between the subject’s (electrogoniometers) and the robot’s (robot’s encoders) joint angles. For the Isometric Condition, we developed a graphical interface with a cursor that tracks the position of the robot’s end-effector, and specific targets that serve as cues to prompt the subject with the isometric contractions to perform. The performance of the algorithm is evaluated using the time needed to accomplish the task. The analysis compares these indexes across conditions (synergy vs muscle-pair methods; Day 1 vs Day 2). Results in 10 able-bodied subjects show that our method is able to provide online, simultaneous, and accurate control of 2 DOFs of a robotic arm. Our work shows the effectiveness of the synergy-based myoelectric control with respect to the traditional single-muscle approach. Results also report the robustness of the method over days without the need for a daily calibration.