praneethnamburimit/immersionlab-pe-mis-table-wiping

Table-wiping multimodal teaching dataset

A small, fully synchronized multimodal recording of a hand wiping a table, built for the MIT Professional Education course Mastering Integrated Systems: Imaging, Machine Learning & Physical AI. One subject, one continuous trial with four task segments -- Normal / Tensed / Slow / Fast.

Modalities (all on the OptiTrack clock; t=0 is the first mocap frame)

• OptiTrack motion capture (200 Hz) -- right-arm markers.
• Delsys EMG -- right-forearm flexors + extensors (RMS envelopes).
• Telemed ultrasound -- transverse forearm; tissue tracked with DUSTrack (two points, DLC + LK-RSTC). The inter-point distance is the tissue-deformation metric. Per-frame timing comes from the device FrameOutput pulses digitized on the Delsys (~133 fps).
• ATEM video clips (one per condition) with de-hummed audio.

Files

• table_wiping.h5 -- the bundle: groups emg (flexor/extensor RMS), mocap (markers + hand speed), us (inter-point distance, track/point0|point1, frame_times_ot), segments (per-condition OT-time windows). Everything on the OptiTrack clock.
• manifest.json -- channel / segment / fps metadata.
• clips/ -- Normal / Tensed / Fast task clips (de-hummed audio).
• figures/ -- the EMG -> ultrasound -> motion reveal + the condition contrasts.

Load

import h5py, numpy as np
from huggingface_hub import snapshot_download
d = snapshot_download("praneethnamburimit/immersionlab-pe-mis-table-wiping", repo_type="dataset")
with h5py.File(f"{d}/table_wiping.h5", "r") as h:
    t  = h["emg/RForearmFlexors/t_ot"][:]      # OptiTrack seconds
    fe = h["emg/RForearmFlexors/rms_mV"][:]
    us = h["us/distance"][:]                    # tissue deformation (px)

Contrasts

Normal -> Tensed: flexor EMG x2.7 + extensor x2.3 (co-contraction). Normal -> Fast: hand speed x1.4, cadence up. Ultrasound deformation: Tensed > Fast > Normal.

Code + the Day-1 teaching notebook: https://github.com/praneethnamburi/immersionlab-pe-mis