Real-Time Decoding of Movement Onset and Offset for Brain-Controlled Rehabilitation Exoskeleton
Kanishka Mitra, Satyam Kumar, Frigyes Samuel Racz, Deland Hu Liu, Ashish Deshpande, José del R. Millán
AI summary
Problem
Most robotic rehabilitation systems act mechanically rather than neurally, limiting activity-dependent plasticity, while existing BCI controllers struggle with signal drift and movement artifacts during online start-stop control.
Approach
The authors developed an online Riemannian EEG decoding pipeline paired with a class-agnostic fixation-based recentering method to track drift and maintain decoder accuracy during real-time exoskeleton control.
Key results
- First online demonstration of dual-state motor imagery start/stop control of an exoskeleton
- Group-mean hit rates of 61.5% for onset and 64.5% for offset across two sessions
- Identification of class-driven bias in task-based recentering and introduction of fixation-based recentering
- AUC gains of +56% for onset and +34% for offset after drift correction
Why it matters
Provides a practical, intention-contingent control framework that aligns robotic assistance with neural intent to enhance neuroplasticity and clinical translation for stroke rehabilitation.
Abstract
Robot-assisted therapy can deliver high-dose, task-specific training after neurologic injury, but most systems act primarily at the limb level—engaging the impaired neural circuits only indirectly—which remains a key barrier to truly contingent, neuroplasticity-targeted rehabilitation. We address this gap by implementing online, dual-state motor imagery control of an upper-limb exoskeleton, enabling goal-directed reaches to be both initiated and terminated directly from noninvasive EEG. Eight participants used EEG to initiate assistance and then volitionally halt the robot mid-trajectory. Across two online sessions, group-mean hit rates were 61.5% for onset and 64.5% for offset, demonstrating reliable start–stop command delivery despite instrumental noise and passive arm motion. Methodologically, we reveal a systematic, class-driven bias induced by common task-based recentering using an asym- metric margin diagnostic, and we introduce a class-agnostic fixation-based recentering method that tracks drift without sampling command classes while preserving class geometry. This substantially improves threshold-free separability (AUC gains: onset +56%, p = 0.0117; offset +34%, p = 0.0251) and reduces bias within and across days. Together, these results help bridge offline decoding and practical, intention-driven start–stop control of a rehabilitation exoskeleton, enabling pre- cisely timed, contingent assistance aligned with neuroplasticity goals while supporting future clinical translation.