Wearable Extra Robotic Arms: Data-Driven Control for Supernumerary Robotics Arms

While extra robotic limbs hold promise for enhancing human capabilities through physical assistance, challenges persist in improving their effectiveness, cooperative control, safety, and overall user-friendliness. This study developed an integrated system using wearable extra robotic arms (ERAs), soft grippers, and glove sensor interfaces to enable shared control of complex cooperative manipulation tasks. Lightweight 4-DOF ERAs provided dexterous reaching assistance, while soft grippers employing pneumatic actuation permitted gentle object grasping. A customized sensor glove incorporating flex sensors and an inertial measurement unit (IMU) was applied to wirelessly measure the user's hand movements. A machine learning approach was implemented for coordinated control, in which the user's hand motion angles measured by the glove sensor drove the extra arms through a neural network model trained on paired human–robot arm data. The contribution of this work lies in embedding a lightweight neural network into a constrained microcontroller to achieve real-time proprioceptive mapping between the user’s biological motion and the robotic limbs. This real-time biological–robotic arms mapping directed cooperative motions without explicit control programming or inverse kinematic computation. Seven multi-arm cooperative object manipulation tasks were conducted over repeated trials. The results confirm the feasibility of the proposed wearable extra robotic arms equipped with a machine-learning interface in assisting users with complex manipulation tasks that are challenging for a single person to perform.