Overview This dataset contains raw recordings of surface electromyography (sEMG) and inertial measurement unit (IMU) data collected during ankle exoskeleton experiments. The data is organized by action, subject, and session (part), stored as individual CSV files.

Processing code is publicly available at: https://github.com/Sr933/exoskeleton

/RawData/ └── ActionName1/ ├── SubjectA/ │ ├── Part1/ │ │ ├── file1.csv │ │ ├── file2.csv │ │ └── ... │ └── Part2/ │ └── ... └── SubjectB/ └── ...

ActionNameX: e.g., Walking_forwards, Pick_up_object, Turn_left, etc. SubjectX: unique subject identifier (e.g., S01, S02, …). PartX: session or trial split identifier (e.g., Part1, TrialA, etc.). fileN.csv: raw data file for a given trial, as captured from the sensors. CSV File Contents Each CSV contains time-series data including:

Time/Index columns (e.g., time, timestamp, idx, id) Raw sEMG channels (integer ADC readings) – 8 total: EMG1–4 (right leg): Tibialis Anterior, Gastrocnemius Medialis, Gastrocnemius Lateralis, Soleus EMG5–8 (left leg): Tibialis Anterior, Gastrocnemius Medialis, Gastrocnemius Lateralis, Soleus Raw IMU channels – at least 3: IMU1: Right shank IMU2: Left shank IMU3: Right foot Channel names may appear next to corresponding data columns and are typically assigned as EMG1, EMG2, ..., IMU1, etc.

Sensor-to-Muscle Mapping EMG1–EMG4 (right leg): Tibialis Anterior, Gastrocnemius Medialis, Gastrocnemius Lateralis, Soleus EMG5–EMG8 (left leg): Tibialis Anterior, Gastrocnemius Medialis, Gastrocnemius Lateralis, Soleus IMU1: Right shank IMU2: Left shank IMU3: Right foot

Sensor Groupings (for analysis convenience) Right leg = IMU1 (shank) + IMU3 (foot) + EMG1–4 Left leg = IMU2 (shank) + EMG5–8 Processing Pipeline (as provided in the GitHub repository) Read CSV files grouped by action → subject → part.

Preprocessing: EMG: ADC-to-voltage conversion, Hampel filter for outlier suppression (window ≈ 50 samples), band-pass filtering (0.2–400 Hz, 5th-order Butterworth), z-score normalization. IMU: Band-pass filtering (0.2–10 Hz, 5th-order Butterworth), z-score normalization.

Metadata extraction: Each processed file results in a struct with subject, part, action, and filename, along with the data table.

See the full MATLAB preprocessing scripts here: https://github.com/Sr933/exoskeleton