Most velocity sensors on the market rely on a single-axis assumption: one reference frame, one direction of travel, one fixed point of measurement. That approach holds up reasonably well when the load travels in a perfectly straight line. It breaks down the moment it doesn't, and in practice, almost no barbell trajectory is a clean straight line. Even a "strict" vertical pull carries lateral and rotational components that a single-axis model simply cannot resolve.
When I designed ORCA's sensor architecture, I built it around that constraint rather than around it. I settled on two 6-axis IMUs, cross-referencing each other at 200 samples per second, to reconstruct the full 3D path of the bar. The reasoning behind that decision is worth unpacking.
The single-sensor problem
A 6-axis IMU reports linear acceleration on three axes and angular velocity on three more. That's a rich dataset, but a single unit has an intrinsic limitation: it cannot fully disambiguate rotation from translation when both occur simultaneously, which is the default condition for almost any compound lift. A bench press, a pull, a squat all involve some combination of angular displacement and linear travel that a lone sensor conflates.
The deeper issue is what happens downstream. Velocity is obtained by integrating acceleration over time, and integration is unforgiving: any small measurement error compounds with every subsequent sample. This is the classic drift problem. A single-IMU system can report a plausible-looking number for the first few reps of a set, while the underlying estimate is diverging further from the true kinematics with every rep. I ran into this early in development, and it's what pushed me away from a single-sensor design in the first place.
The fix: sensor fusion, not sensor addition
I placed two IMUs at distinct points on the same rigid body and cross-referenced their outputs in real time. This isn't just redundancy. It's a qualitatively different measurement problem. With two spatially separated reference points, I can:
- Disambiguate rotation from translation with far greater confidence, because a shared rotational component appears differently at each sensor location, which makes it mathematically separable from pure translation.
- Bound drift before it accumulates, using each IMU as a real-time check on the other rather than trusting a single, unconstrained integration path.
- Reconstruct the true 3D bar path: not merely a velocity projected onto an assumed axis, but the actual trajectory through space, whether strict, curved, rotated, or fully free.
This is the architecture that lets me hold the same accuracy on a strict bench press and a sumo pull that drifts noticeably off vertical, two movements that a tethered cable system or a camera-based system would handle very differently, and inconsistently.
Why this matters for velocity-based training
A load-velocity profile is only as good as the consistency of the velocity measurement underneath it. If accuracy degrades depending on bar path angle, every downstream decision (target velocity zones, velocity-loss thresholds for autoregulation, load adjustments) is built on a signal that shifts without anyone noticing. That's not a rounding error; it's an unmodeled confound baked into the training data, and it's exactly what I set out to eliminate at the sensor level rather than patch over in software.
Two-sensor fusion removes that variable. The measurement holds whether the bar travels on a perfectly vertical path or deviates by several centimeters, which in a real training environment is the norm rather than the exception.
The practical implication
For a coach or athlete, the engineering translates into something simple: a single device, magnet-mounted to any bar in about a second, delivers accuracy comparable to a floor-tethered reference system, without a cable, without recalibration between racks, and without being restricted to strictly vertical movements.
That's the underlying design principle I followed at every stage of building ORCA: push the sensing problem hard enough at the engineering layer that the constraint disappears for the end user.
Léo Fortin Dionne designed the advanced measurement systems behind ORCA Strength Systems.