Velocity-Based Training: 1RM From Bar Speed

A 1RM that took an hour and three near-misses to establish is stale by the next training session. Sleep, food, stress, and inflammation all move maximal output around 5 to 10 percent week to week. Velocity-based training (VBT) replaces the periodic max-out with a single warm-up set: load the bar, lift one rep at maximum intent, read the bar speed off a linear position transducer or a phone-based optical sensor, and the calibrated load-velocity profile gives you today's estimated 1RM.

The math behind VBT is the load-velocity profile: the linear relationship between percentage of 1RM and mean propulsive velocity at that load. Gonzalez-Badillo and Sanchez-Medina established the bench-press version in 2010 with eighty trained men and a regression that explained ninety-eight percent of the variance.^[1] Subsequent work extended the model to the squat, deadlift, bench pull, and Olympic-lift variants. This is the math, the calibration procedure, the failure modes, and a worked squat example.

The load-velocity profile

Each exercise has its own linear mapping from percent of 1RM to mean propulsive velocity. The general form:

%1RM = a × MPV + b

  MPV = mean propulsive velocity (m/s)
        average velocity during the propulsive phase
        of the concentric action (Sanchez-Medina 2010)
  a, b = exercise-specific regression coefficients

For the bench press, Gonzalez-Badillo and Sanchez-Medina fit the population-level coefficients on a sample of 80 trained men (resistance-trained, mean 1RM 1.04 × bodyweight). The regression: %1RM = -90.43 × MPV + 119.22, valid in the 30 to 100 percent 1RM range.^[1] Read in the other direction: at MPV = 0.50 m/s, the predicted %1RM is 119.22 - 90.43 × 0.50 = 74.0 percent.

The coefficients change by exercise because the lever arms, ranges of motion, and minimum velocity at 1RM all differ. Sanchez-Medina, Gonzalez-Badillo, Perez, and Pallares fit the bench-pull version in 2014 and reported separate coefficients (and a separate minimum velocity).^[2] Banyard, Nosaka, and Haff did the back squat in 2017 with 17 trained males and concluded that individualised profiles outperform the group mean.^[3]

Minimum velocity threshold

Every lift has a velocity floor at 1RM: the lift completes only because the neuromuscular system finds the last drop of force at the slowest possible bar speed. This floor is the minimum velocity threshold (MVT). Reported population means:

• Bench press: 0.17 m/s (Sanchez-Medina, group mean from the 2010 bench paper).^[1]
• Back squat: 0.30 m/s (Banyard 2017; individualised range 0.23 to 0.37 m/s).^[3]
• Deadlift: ~0.15 m/s. Slower because the lockout phase is heavily decelerative; individual variation is high.
• Bench pull: 0.51 m/s (Sanchez-Medina 2014; the longer concentric range and absence of a deceleration phase shifts the floor up).^[2]
• Power clean / snatch from blocks: 0.7 to 1.0 m/s. These are velocity-dominant lifts; the bar must move fast enough to finish in the catch position.

The MVT is the velocity at which the load-velocity regression intersects the 100 percent 1RM line. It is an exercise property, not a training property. It does not improve with training. What improves is the load lifted at that velocity.

Why mean propulsive velocity, not peak velocity

Sanchez-Medina, Perez, and Gonzalez-Badillo argued in 2010 that mean propulsive velocity (MPV) is the cleanest velocity variable for load-velocity profiling.^[5] Two reasons. First, peak velocity occurs at different points in the lift across loads; at 90 percent 1RM there is no meaningful peak above the propulsive average. Second, the deceleration phase of light loads (above ~0.5 m/s) systematically biases mean concentric velocity downward and adds noise. MPV is computed only across the portion of the concentric where the bar is still accelerating against gravity.

Most commercial VBT devices (Vitruve, GymAware, Tendo, Open Barbell) report both MPV and mean velocity. For load-velocity profiling, use MPV. For velocity-loss tracking inside a set, mean velocity is acceptable because the comparison is within the same load.

Calibration: building your own profile

Population profiles are decent for trained men in the 80 to 95 kg bench range. Individuals vary, especially at the squat. Banyard 2017 reported that individualised profiles produced 1RM estimates within 5 percent for 14 of 17 squatters, while the group profile missed by 8 to 12 percent for half the cohort.^[3]

The calibration protocol:

1. Pick four to six loads spanning roughly 30 to 90 percent of estimated 1RM. For a squatter with a known 180 kg max, that's 60 / 90 / 120 / 140 / 160 kg.
2. For each load, perform two to three reps with maximal intent. Record the fastest MPV.
3. Plot %1RM (computed from the known 1RM) against MPV. Fit a linear regression.
4. Solve for the coefficients a and b. Solve for MVT by setting %1RM = 100 and reading off MPV.

Re-calibrate after a major training block (8 to 12 weeks) or after a deload-and-test cycle. The coefficients drift slightly as muscle architecture and rate-of-force-development change, but the structure is stable.

Worked example: squat 1RM from a single warm-up set

A 90 kg lifter has a calibrated squat profile fit on five loads:

%1RM = -110 × MPV + 130
  (individualised; Banyard-style protocol, 5 loads, R^2 = 0.97)
  MVT = 0.27 m/s (intersect with %1RM = 100)
  Estimated 1RM (last test): 175 kg

Today's warm-up
  Load:  140 kg (80% of last test)
  MPV:   0.55 m/s   (mean of 2 reps, fastest)

Predicted %1RM at this load
  %1RM = -110 × 0.55 + 130 = 69.5%

Predicted 1RM
  1RM   = 140 / 0.695 = 201 kg

Reality check
  Last test 1RM:        175 kg
  Predicted today:      201 kg
  Delta:                +15% over last test

Honest interpretation
  Either the lifter is genuinely 10–15% stronger than the last test
  (plausible after an 8-week block), or today's warm-up MPV at 140 kg
  is unsustainably fast (over-priming, fresh CNS, low fatigue baseline).
  Use the predicted 1RM as today's anchor for top sets, not as a new max.

The point of VBT is not to inflate a max. It is to set today's working weights at a calibrated %1RM rather than a stale one. If the warm-up at 140 kg comes in at 0.55 m/s, the 80 percent set should be programmed at the current 80 percent of estimated 1RM (160 kg, not 140), and the lifter learns from one warm-up set what they would otherwise discover by missing a top single.

Velocity loss as a fatigue marker inside the set

Sanchez-Medina and Gonzalez-Badillo 2011 used the same load-velocity model to track within-set fatigue. They showed that the percentage drop in MPV from the first rep to the last rep correlates near-linearly with metabolic stress (lactate) and electromyographic markers of fatigue.^[4] A 20 percent velocity loss is associated with moderate metabolic accumulation; 40 percent velocity loss represents substantial neuromuscular fatigue and a longer recovery cost.

Pareja-Blanco and colleagues turned this into a programming variable in 2017.^[7] They compared 8-week training blocks that prescribed 20 percent vs 40 percent velocity loss as the set-termination cue. Both groups gained strength. The 20 percent group preserved more sprint and jump performance; the 40 percent group accumulated more hypertrophy. Pareja-Blanco 2020 extended the result and showed that velocity loss greater than 25 percent confers diminishing strength returns per unit fatigue.^[10]

Practical rule: cap velocity loss at 20 percent for strength and power-focused blocks, allow up to 30 to 40 percent for hypertrophy phases. Read the cap from the device or call the set when the rep speed visibly drops one tier below the opening rep.

Hardware: what's accurate enough

Linear position transducers (GymAware, Tendo) are the reference standard. Reliability against 3D motion capture is better than 2 percent error.^[12] Optical sensors (Vitruve) and phone-camera apps come within roughly 4 percent on bench and squat. Balsalobre-Fernandez 2017 validated an iPhone app against a Tendo unit and reported intraclass correlation greater than 0.95.^[8] Banyard 2017 compared four methods on the back squat and found that linear position transducers and motion capture agreed within 1 to 2 percent, while accelerometer-based units drifted up to 6 percent at the slow end of the curve.^[11]

For load-velocity profiling, accuracy at the slow end (less than 0.4 m/s) matters more than at the fast end. A 0.05 m/s error at MPV = 0.20 m/s shifts the predicted %1RM by 5 to 6 percentage points. The same error at MPV = 1.0 m/s shifts the prediction by less than 1 point. Cheap accelerometer units are fine for tracking velocity loss inside a set; for 1RM estimation, prefer a transducer or a validated optical sensor.

Where the model breaks

• Untrained lifters. The Pareja-Blanco 2014 sample included only resistance-trained men.^[6] Untrained lifters have unstable form at sub-maximal velocities and high week-to-week variance in MVT. Calibration error at 1RM is 10 to 15 percent in the first 8 weeks of training.
• Technique-bound lifts. The Olympic lifts and the deadlift have wider individual variation in MVT because the path of the bar and the catch position change the velocity floor. Population profiles miss by 5 to 8 percent.
• Maximal-effort grinds. A true 1RM at 0.17 m/s on bench involves a 4 to 6 second concentric. Untrained lifters can't grind that long. The MVT they actually fail at is closer to 0.25 m/s, which inflates the predicted 1RM. Calibrate against a real grind, not a missed lift.
• Devices that report mean velocity, not MPV. Mean velocity at light loads (above 0.6 m/s) is biased by deceleration and produces a lower %1RM estimate than MPV. Convert via the device manufacturer's documented offset, or stick to MPV for the profile.

Programming: VBT-prescribed loads

A typical VBT-prescribed session reads loads from velocity, not from a stored 1RM:

Squat day, hypertrophy phase
  Target velocity range: 0.55 to 0.65 m/s (~70-80% 1RM)
  Velocity-loss cap:     30%
  Sets:                  4
  Rep cap:               12

Warm-up
  Bar  × 8     (form)
  60kg × 5     0.95 m/s
  100kg × 3   0.78 m/s
  140kg × 1   0.62 m/s   ← within target range, this is the working load

Working sets
  Set 1: 140kg × 8 reps until MPV drops to 0.43 m/s (30% loss from 0.62)
  Set 2: 140kg × 7 reps until MPV drops to 0.43 m/s
  Set 3: rest extended; rep count drops naturally
  Set 4: lower load 5kg if rep count is below 5

The lifter never tests a 1RM. The calibrated profile and the device take care of the percentage. On a fresh day, 140 kg sits at the bottom of the target range and the working sets run long; on a fatigued day, the same 140 kg comes up at 0.55 m/s and the sets terminate sooner. The training intent is preserved; the load adapts.

Cross-link tools

• 1RM Calculator uses Epley, Brzycki, and Lombardi rep-based formulas and is the comparison point for VBT-derived 1RM estimates.
• RPE-to-Percentage Converter covers the alternative auto-regulation framework for lifters without VBT hardware.
• Strength Standards places a calibrated 1RM in population context.

• VBT estimates today's 1RM from a sub-maximal warm-up set using a calibrated linear load-velocity profile.
• The math: %1RM = a × MPV + b, with exercise-specific coefficients and a minimum velocity threshold near 0.17 m/s (bench) or 0.30 m/s (squat).
• Population profiles are decent for trained lifters; individualised profiles outperform group means by 3 to 5 percent.
• Velocity loss inside a set tracks neuromuscular fatigue. Cap at 20 percent for strength, 30 to 40 percent for hypertrophy.
• Linear position transducers are the reference standard; validated optical and phone-based sensors are within 4 percent.
• The model breaks for untrained lifters, technique-bound lifts, and devices reporting only mean velocity.