Velocity-based training for powerlifting, with two sensors on the barbell.
KTH Sports Technology HT26 · Arvid · Greipur · Hugo · Silje · Svanberg

---

V-LO measures how fast the barbell moves on every rep, using two Movesense IMUs on the bar ends, and turns it into feedback for the next set. Instead of a fixed weight, the coach gives each lift a target velocity zone; the athlete records the set from the browser, and a few seconds after pressing Stop sees the mean and peak velocity of each rep, whether the set hit the zone, and a suggested load for the next set.

This repository holds the web platform, the database setup and the cloud processing. The velocity algorithm is developed by the data team in Product-testing and copied in unchanged.

How it works

1. Sensors: two Movesense sensors stream accelerometer and gyroscope data at 104 Hz over Bluetooth, straight into the browser (Web Bluetooth).
2. Web app (vlo.html, js/): records the set and uploads the raw data with the set's reps, load and target zone.
3. Supabase (PostgreSQL + storage): stores the raw JSON in raw-sessions and a row in sessions. A database webhook starts the processing; a scheduled job retries any recording that was missed.
4. Azure Functions (python/function_app.py): runs the data team's analysis (python/squat/, via python/squats.py) and saves the velocity of every rep back to the database.
5. Feedback: the app shows the result as soon as it is saved, with no refresh.

Details: docs/azure-processing.md · docs/database.md · next steps

Running it locally

The website is plain HTML, CSS and JavaScript; any static file server works:

python -m http.server 8080

Then open http://localhost:8080/vlo.html in Chrome or Edge (Web Bluetooth). Log in with your V-LO account. Adding ?mirror=1 lets a single sensor stand in for both, for testing only.

The analysis can also be run on a laptop for debugging, e.g. python main.py --session <id> from the python/ folder; see the top of python/main.py. It needs numpy, scipy, pandas, matplotlib and imufusion (python/requirements.txt).

Repository

Path	What
vlo.html, js/, css/	The web platform: training, plans, teams, recording
python/	Cloud processing (Azure Functions) and the local debugging tool main.py
python/squat/	The data team's squat analysis, copied unchanged by python/sync_data_team.py
supabase/	Database schema, migrations and read-only queries
docs/	Architecture, database and processing documentation

Future work

• Better drift handling to make fast, light lifts as accurate as heavy ones
• Bench press and deadlift analysis, validated like the squat
• A larger validation study with powerlifters
• Combine both sensors, e.g. to measure left–right balance
• Automatic rep detection, so no rep count has to be entered
• Estimate the 1RM from the athlete's load–velocity profile