EMGNet: An EMG Dataset for Locomotor Intent Recognition

Surface electromyography (EMG) can be used to interact with and control robotic systems via intent recognition. However, most machine learning algorithms for EMG intent recognition have been trained using small-scale data with limited individuals, which can affect generalization across users and tasks. Motivated by these limitations, we developed a large-scale EMG dataset to support research and development in intent recognition systems, with an emphasis on human locomotion. Our new dataset combines multiple open-source datasets, as outlined in section 3, with processed EMG signals for healthy subjects. Each dataset that we included in our meta-dataset has been modified to achieve a consistent data structure for ease of use and to establish a standardized pipeline for data preprocessing (e.g., filtering, normalization, and windowing) and for training machine learning algorithms. Our EMG dataset is 152 GB comprised of 7 open-source datasets with 132 users total from four different countries. Signals include tibialis anterior, medial gastrocnemius, rectus femoris, and biceps femoris, and six activity classes, including standing, level-ground walking, stair ascent, stair descent, ramp ascent, and ramp descent.