Self-derived Motion Features from sEMG for Inferring 3D Forearm Trajectories

Surface electromyography (sEMG) records the activation patterns of muscles. It has been widely studied for human motion estimation in wearable robotics and human-machine interaction. Most existing methods focus on mapping sEMG signals to intermediate kinematic variables (joint angles, torque, or intended motion). Inferring 3D trajectories directly is challenging, as it generally requires additional motion features from other sensing modalities. Unlike the previous works, we proposed a method to derive extensible motion features from sEMG to infer 3D motion trajectory. To validate, an in-vivo experiment was performed to collect nine different typical forearm motions for ten subjects. The results showed that the self-derived motion features improved the motion estimation performance by an average of 10% to 15% across evaluation metrics. This revealed the closed relationships of the derived motion features, and their contributions to the 3D trajectory. The results showed the potential for deriving more critical motion features from sEMG within a single framework, which assists in an in-depth understanding of forearm motor control and improves the human motion trajectory inference in future robotics applications.