A Hierarchical Machine Learning Approach for Real-Time BMI Control of an Upper-Body Exoskeleton

Kanishka Mitra1,2, Frigyes Samuel Racz1,3, Anna Bucchieri2, Satyam Kumar1, Hussein Alawieh1, Ashish D. Deshpande2, José del R. Millán1,3
1 Chandra Family Department of Electrical and Computer Engineering, The University of Texas at Austin 2 Walker Department of Mechanical Engineering, The University of Texas at Austin 3 Department of Neurology, The University of Texas at Austin IJCAI Workshop Demo, 2023
Paper Video

Abstract

In recent decades, human-robot interaction and brain-machine interfaces have both advanced as tools for neurorehabilitation, but their integration remains largely unexplored. This workshop demo presents a real-time hierarchical machine-learning approach that detects the onset and offset of motor imagery to control passive reaching with an upper-body exoskeleton. Instead of aiming for continuous control, the system allows a more natural sense of functional interaction by initiating movement when motor imagery begins and terminating movement when motor imagery ends, while remaining robust to exoskeleton-induced motion and noise.

Demo Video