TL;DR
- ▸Human motion research fuses many sensor streams, motion capture, inertial units, force plates and high-speed video, and every one of them stamps its data with its own clock.
- ▸When those clocks disagree by even a few milliseconds, the fused picture is wrong: a foot strike lands in the wrong video frame and the force reading is attributed to the wrong moment of the gait cycle.
- ▸Disciplining every capture device to one shared reference removed the alignment guesswork, made experiments reproducible, and let the group trust derived measurements they previously had to throw away.
The research, and the hidden variable underneath it
The biomechanics group at Atlantic Technological University studies how people move: gait, balance, sports technique, and rehabilitation after injury. To do that they record the same movement through several instruments at once. Optical motion capture tracks reflective markers in three dimensions. Inertial measurement units on the limbs report acceleration and rotation. Force plates in the floor measure ground reaction forces. High-speed cameras capture the visual ground truth. Each instrument is excellent on its own.
The science only works when those streams are combined. A researcher wants to say that at the instant the heel struck the ground, the knee was at a particular angle and the force plate read a particular load. That sentence contains the word instant, and instant is where multi-modal sensing quietly falls apart. Every device timestamps its samples against its own internal clock, and those clocks are not the same clock.
The core problem
Sensor fusion is really clock fusion. If the devices do not agree on what time it is, the data cannot be aligned after the fact without guesswork, and the guesswork is where reproducibility goes to die.
Why a few milliseconds matters here
Human movement is fast. During running, the foot is in contact with the ground for around a fifth of a second, and the most interesting biomechanics happen in the first few tens of milliseconds of that contact. A high-speed camera running at 1000 frames per second produces a frame every millisecond. If the force plate clock and the camera clock differ by five milliseconds, the force peak is attributed to a video frame five frames away from where it actually happened.
For a casual demonstration that does not matter. For research that has to survive peer review, it is fatal. A derived measurement built on misaligned inputs is not noisy, it is wrong, and worse, it is confidently wrong in a way that looks plausible. The group had been spending real effort manually nudging streams into alignment using clapboard-style sync events, and every manual alignment was a source of error and an obstacle to reproducing a result months later.
The approach: give every device the same clock
The fix is conceptually simple. Instead of letting each instrument keep its own time and trying to reconcile afterwards, you discipline every capture device to one shared time reference, so that a timestamp from the force plate and a timestamp from the camera mean the same thing.
The capture machines run the synchronisation agent, which takes control of the host clock and aligns it to a common grandmaster reference over the lab network using the Precision Time Protocol. Devices that can consume PTP directly lock to the same reference. Devices that cannot are driven from a host that can, so their samples are stamped against a disciplined clock rather than a free-running one. The result is a single laboratory timebase that every stream is measured against.
Nothing about the sensors changed. The motion capture system, the force plates and the cameras are the same hardware the group already owned. What changed is that their data now lands on a shared timeline instead of several drifting ones.
What changed for the researchers
The first change was that alignment stopped being a manual step. Recordings from a session dropped onto a common timeline automatically, so a heel strike in the video sat in the same instant as the force it produced and the joint angle that accompanied it. The clapboard sync ritual went away.
The second change was reproducibility. Because the timeline is defined by a disciplined reference rather than by a researcher lining up streams by eye, the same raw recording produces the same fused result every time it is processed, by any member of the group, months apart. That is the difference between a demonstration and an experiment.
The third change was trust in derived measurements. Quantities that depend on cross-stream timing, joint power, loading rate, time-to-stabilisation, were previously treated with suspicion because the alignment underneath them was shaky. With a shared clock those quantities became dependable enough to build conclusions on.
The outcome in one line
The sensors did not get better. The time did. And once the time was trustworthy, every measurement built on top of it became trustworthy too.
Why this generalises beyond one lab
Atlantic Technological University is a clean example of a pattern that shows up everywhere multiple instruments observe the same event: autonomous vehicle sensor suites, industrial test rigs, telemetry in motorsport, distributed scientific instruments. The instinct is to invest in better sensors. Often the real limit is not sensor quality at all, it is the assumption that independent clocks can be reconciled after the fact.
Any time you fuse data from more than one device and you care about the order or the coincidence of events, time synchronisation is not an accessory to the measurement, it is part of the measurement. Treating it that way, from the start, is far cheaper than discovering it after a dataset has to be discarded.
Frequently asked questions
Why not just align the streams in software afterwards?+
What accuracy is actually needed for human motion research?+
Did the lab have to replace its sensors?+
Related reading
Blog · Positioning
When Milliseconds Are Not Good Enough
For most of computing, millisecond-class clock synchronisation is fine. For a growing list of use cases — finance, broadcast, 5G, AI, distributed databases — it isn't. A field guide to recognising when your application has crossed the line and millisecond-grade time has become a liability.
Blog · Protocols
Precision Time Protocol vs NTP: When Each Belongs in Production
The honest engineering comparison between Precision Time Protocol and NTP — what each protocol can actually deliver, where the boundary lives, and how to choose between them without falling for either side's marketing.
Blog · Standards
Understanding IEEE 1588 PTP: How Precision Time Powers Industrial Ethernet
What IEEE 1588 actually defines, how the protocol works at the message level, and why it's the foundation under every modern industrial Ethernet, telecom and broadcast timing fabric.