Data¶
AJ's recordings from 2026-10-22 are the
dataset-2026-10-22 release
(AJ-myo-lever-dataset.zip, about 3.7 MB). One driver, right arm, two wearings of the band.
| Folder | What it holds |
|---|---|
session1/, session2/ |
Six poses held for 2, 3 and 5 s: Myo_EMG (8 channels, 200 Hz) and Myo_IMU (10 channels, 50 Hz) per take |
results/runs/ |
The sim's official run CSVs: score, collisions, gates, smoothness, workload |
results/drives/ |
Every controller drive log: one row per 20 Hz command, with the EMG features the controller saw |
results/blind/ |
The keys for the blind lag-compensation test |
File names give the pose:
| Pose | EMG label | IMU label |
|---|---|---|
| GO (arm forward, relaxed) | rest |
neutral |
| STOP (arm raised) | arm_up |
arm_up |
| LEFT / RIGHT (forearm rolled) | roll_left / roll_right |
roll_left / roll_right |
| REVERSE (arm tipped down) | tip_down |
tip_down |
| BRAKE (firm fist) | forward |
brake |
Use it¶
python examples/myo_gestures.py drive <path>/session2 --mode lever # drive with AJ's calibration
python examples/team_results.py <path>/results # recompute the results tables
In Python, gestures.load_recordings(folder, "Myo_EMG", labels, 8) from the repo's examples/ loads a
folder. Split train and test by session or by person, never by window: windows from one take are
near-copies of each other.
Other drivers¶
The four-driver comparison on the Results page also used recordings from Jose, Kiran and Xinyi. This site publishes only AJ's recordings and summary figures from the others.