How accurate is ODKI-VBT?
Vertical bar movement, per rep.
The sensor tracks motion along the vertical axis in the world frame. For each rep it reports peak and mean velocity, peak and mean acceleration, displacement, plus peak and mean velocity of the eccentric phase.
- Measured
- Vertical velocity, acceleration and displacement, per rep
- Not measured
- Horizontal bar path, bar tilt, mean propulsive velocity
- Mounting
- Magnets on a steel bar; non-magnetic bars won't hold it
- Firmware
- v3.11.31
The IMU is read at 416 Hz and averaged to 100 Hz. A Madgwick filter estimates orientation, so acceleration can be rotated into the world frame and integrated into velocity. Velocity is never force-reset: drift is measured during brief moments of stillness and removed retroactively, so a rep's numbers can be refined after it ends. Every stage is documented in the firmware docs ↗.
A self-check
It compares the rep's integrated displacement with its mean velocity × duration. A low score flags a rep worth a second look. It is a signal-consistency score, not an accuracy estimate: a uniform scale error, for example, would leave it at 100. Whether it predicts the error will be tested in the validation.
Final or not yet
Each rep says whether its numbers are provisional or already drift-corrected and final.
Reproducible
Over USB, every sample is logged at 100 Hz. A Python viewer and a replay tool that runs the real tracking engine are in the repo ↗.
Compared with a linear encoder.
A summary. The full protocol, with the acceptance thresholds fixed before any data is collected, is in the public repo.
| Reference | A linear encoder attached to the same bar. |
|---|---|
| Checking the reference | The encoder is itself checked against a tape measure before it's used, with the limits fixed in the protocol. |
| Devices | 2 ODKI-VBT units, mounted next to the encoder on the same side of the bar. |
| Lifts and loads | Back squat and bench press, from light to heavy loads (percentages of an estimated 1RM, set in the protocol; no maximal test). |
| Statistics | Per rep: Bland–Altman bias and limits of agreement, ICC, SEM, CV and mean absolute error; plus reps detected, missed or invented. Both the value shown live and the final drift-corrected value are compared. |
| Wording | Results with our encoder are reported as "agreement with a linear encoder". "Validated" is used only after co-validation with a university lab, planned for 2027. |
- Check the reference encoder.
- Technical pilot: checks that the pipeline works. Its numbers are not published.
- Test with a small group of lifters, with written consent and anonymous IDs.
- Publish the report, the dataset and the scripts.
- Co-validation with a university lab using the same protocol, planned for 2027.
- The paired raw data, ODKI and encoder, for every rep
- The analysis scripts, so anyone can rerun them
- Results by lift and by velocity range, including where it does badly
- The firmware version used for every number
- An open dataset under CC BY 4.0
No results date yet: we'll add one only when we can keep it. Until results are out, this page and the rest of the site don't describe ODKI-VBT as accurate, precise or validated.
Run a lab with an encoder or motion capture?
Co-validate ODKI-VBT with us using the same published protocol. Results are published openly, whatever they show.
- 26 Oct 2026
- First version: status, method summary, what we'll publish.