Metric vs Vitruve LPT: Smith machine bench press at 1-RM (Grossi et al., 2026)
A 2026 peer-reviewed study from the University of Urbino tested MetricVBT against a Vitruve linear position transducer during Smith machine bench press at 1-RM. Peak velocity agreement was good; mean velocity was moderate. Here is what the data shows and what it means in practice.
A new peer-reviewed paper from the University of Urbino compared MetricVBT against the Vitruve linear position transducer (LPT) during Smith machine bench press at 1-repetition maximum.
Grossi, T., Micheli, L., Magnoni, M., Shoaei, V., Benelli, P., Ferri Marini, C., & Lucertini, F. (2026). Comparison of the MetricVBT App and the Vitruve Linear Position Transducer for Assessing Execution Velocity and ROM. Journal of Functional Morphology and Kinesiology, 11(2), 197. https://doi.org/10.3390/jfmk11020197
Study overview
Eighteen resistance-trained men performed a single 1-RM Smith machine bench press (SMBP) session. Barbell kinematics were recorded simultaneously by:
- MetricVBT on an iPhone 12, 720p at 60fps, tripod-mounted 1.5 m lateral to the rack with lens at 1.30 m height — perpendicular to the barbell plane as per the app’s setup guidelines
- Two Vitruve linear position transducers sampling at 1000 Hz, attached to the left side of the barbell
Three metrics were compared for the 1-RM attempt: mean velocity (MV), peak velocity (PV), and range of motion (ROM).
What the study found
Peak velocity was the strongest result. Pearson’s r = 0.91 (very high correlation), ICC = 0.888 (good agreement), mean bias of just +0.02 m/s. The limits of agreement were −0.08 to +0.11 m/s.
Mean velocity was moderate. Spearman’s ρ = 0.65, ICC = 0.612, mean bias −0.02 m/s — near-zero systematic error. Limits of agreement were −0.07 to +0.04 m/s.
Range of motion was poor. ICC = 0.236, mean bias 1.64 cm, limits of agreement −13.82 to +17.10 cm. The authors identified three ROM data points as likely recording errors (two from Metric, one from the Vitruve unit) — when those were excluded in a sensitivity analysis, the limits of agreement improved to −4.26 to +5.51 cm. Similarly, two MV outliers (both from Metric) were identified as recording errors; removing them improved the MV limits of agreement to −0.03 to +0.01 m/s.
The paper’s conclusion is that while agreement and correlation were found, MetricVBT and VitruveLPT should not be considered interchangeable devices based on this dataset, primarily because the limits of agreement are wide relative to the magnitudes being measured at 1-RM.
Why 1-RM bench press is the hardest test
This study tested only the 1-RM attempt — the single slowest repetition in any set, typically in the range of 0.12–0.20 m/s for mean velocity. At those velocities, a tracking difference of 0.04 m/s represents roughly 20–30% of the measured value. That makes 1-RM conditions a more demanding validity test than submaximal VBT work, where velocities are higher and the same absolute error is a smaller percentage of the reading.
Most practitioners use VBT at submaximal loads — monitoring fatigue, autoregulating volume, building load-velocity profiles. Those conditions produce higher velocities where both tracking systems have more signal to work with. The Šagovac study, which tested Metric across 45–75% 1-RM loads (MV range roughly 0.3–1.0 m/s), found correlations of r = 0.93 for MV and r = 0.91 for PV with excellent ICC values.
The Grossi paper is specifically relevant if you are using Metric for 1-RM testing — a narrower use case, and one worth being aware of.
A note on the comparison device
The Vitruve LPT is a tether-based device. The paper notes that the Vitruve software did not allow simultaneous display of mean velocity and mean propulsive velocity (MPV) — so only one could be selected per recording. The study used mean velocity (not MPV) to match the Metric output. This is worth knowing because the load-velocity relationship used in most VBT literature is built on MPV, which excludes the braking phase. MV includes it. Comparing MV between a tether unit and a vision system is technically valid, but practitioners building load-velocity profiles should check which velocity variable their reference device and their tracking app are actually outputting.
Version context
The paper was submitted in March 2026 and tested the version of Metric available at that time. Our internal validation in June 2026 — 95 reps across bench press, deadlift, squat and cleans using iOS v6.9 and Android v1.2 against a GymAware tether unit — returned CCC 0.982 for mean velocity and CCC 0.975 for peak velocity. The accuracy improvements that led to v6.9 were specifically targeted at peak velocity measurement and at video import quality across varied conditions. Whether those changes improve the 1-RM scenario specifically remains to be tested under controlled conditions.
What to take from this
Peak velocity agreement in this paper was genuinely good — ICC 0.888 with near-zero mean bias. Mean velocity was moderate, driven partly by a small number of data points that the authors themselves classified as recording errors. ROM measurement was the weakest metric, consistent with what other studies have found for ROM across smartphone tracking systems.
If you are using Metric for tracking submaximal sets, monitoring fatigue, or building load-velocity profiles, this study does not change the picture from earlier research. If you are specifically trying to pinpoint 1-RM velocities with high individual precision, the limits of agreement here are worth understanding.
The full paper, including Bland-Altman plots and sensitivity analysis, is open access at the link above.