Brain-machine interfaces (BMIs) translate brain signals into motor commands for assistive devices. Despite significant advances, the long-term effects of BMI training on neural adaptation, classifier stability, and individual variability remain poorly understood. We present a multimodal, longitudinal dataset collected from seven healthy participants over nine sessions spanning 15 to 81 days. The dataset includes high-density electroencephalography (EEG), electrooculography (EOG), inertial measurement unit (IMU) data, and exoskeleton state information during BMI control. During the open-loop training phase, participants performed kinesthetic motor imagery (KI) while a remotely controlled exoskeleton executed walking and stopping commands. After the open loop training phase, the system transitioned to closed loop BMI control. For closed-loop control, lower delta band EEG signals were classified using Local Fisher Discriminant Analysis and a Gaussian Mixture Model. The classifier was continuously updated using open-loop data from Sessions#1-5, after which its parameters were fixed. The dataset also includes post-experiment MRI scans from five participants performing KI while viewing themselves walking in the exoskeleton. This report outlines the experimental setup, data collection, and preliminary validation, providing a resource for future BMI research.
Sarkar et al. (Mon,) studied this question.