How can we continuously estimate elbow angle using a single inertial measurement unit (IMU) and four channels of surface electromyography (sEMG) from the upper arm, without collecting any forearm data?
This talk presents a neural network developed to fuse motion and muscle activity signals and continuously estimate elbow flexion and extension. The estimation pipeline is deployed on a Franka 3 robot arm for a hardware demonstration in which the estimated angles drive the robot’s motion.
The demonstration illustrates how this approach could provide a control input for powered upper-limb prostheses and other assistive devices.