This master thesis aims to develop an advanced deep-learning framework to predict continuous finger joint angles using only forearm sEMG signals. Unlike classification-based methods, continuous prediction captures fine-grained movement information and enables smooth, natural control of a robotic hand. The project will focus exclusively on forearm sEMG, leveraging the rich neuromuscular signals that correspond to finger flexors, extensors, and intrinsic motion synergies.
Students will build a complete system that:
-
Records multi-channel forearm sEMG together with ground-truth finger angles (e.g., via motion capture, glove sensors, or IMUs).
-
Develops a real-time signal-processing and deep-learning pipeline to map sEMG signals to continuous joint trajectories.
-
Implements and evaluates real-time interaction by controlling a robotic hand based on predicted movements.
This work addresses one of the key challenges in upper-limb prosthetics and dexterous robotics—how to reconstruct fine hand movements from non-invasive physiological signals. The resulting system supports more natural manipulation, enabling robotic hands to function as intuitive extensions of the human user.