The End of the Motion Capture Suit
For fifty years, recording movement accurately meant covering the body in markers. That era is closing — and what replaces it changes who gets to study movement.
If you have ever seen behind-the-scenes footage of a video game or animated film, you know the image: a performer in a skin-tight black suit covered in small reflective balls, moving through a studio ringed with specialised cameras. This is marker-based motion capture — for half a century, the only way to record human movement with scientific accuracy.
The suit works because the markers solve a hard problem: telling a camera exactly which point of the body it is looking at. Cameras see pixels, not anatomy. The reflective markers give the system unambiguous reference points — this dot is the left elbow, this one the right hip — that can be triangulated across multiple camera views into precise 3D positions.
The problem is everything else about it. The suits are expensive. The studios are expensive. The cleanup is laborious: every capture session produces data full of gaps and confusions that human technicians repair by hand. And — most importantly for anyone studying real movement — the suit changes what it records.
The Observer Effect of the Suit
Ask any dancer who has performed in a motion capture suit: you do not move the same way wearing it. The suit is a costume, the studio is a stage, and the awareness of being recorded — of producing data — alters the quality of attention that somatic practice depends on.
For solo movement, this is a tolerable distortion. For movement involving physical contact between people, it is much worse. Markers attached to two bodies in contact get occluded, knocked off, or confused with each other — the system loses track of which marker belongs to whom precisely at the moments of contact. And those moments are often exactly what the researcher wants to study.
This is why, for example, Contact Improvisation — a practice built entirely on the physics of shared weight and rolling points of contact between bodies — has essentially never been captured at marker-grade accuracy. The capture technology fails precisely where the practice lives.
What Changed This Week
At CVPR 2026, the leading computer vision conference, a system called MAMMA (Markerless Accurate Multi-person Motion Acquisition, from the Max Planck Institute for Intelligent Systems) was presented as one of the conference's top-rated papers. Its claim: motion capture accuracy competitive with commercial marker-based systems, from ordinary multi-view video, with no markers, no suits — and crucially, for multiple people interacting closely.
Two technical moves make this possible.
First, instead of tracking a handful of marker positions, the system predicts dense landmarks — hundreds of points across the entire body surface — directly from video, using a transformer network trained to recognise body geometry. Where a marker suit gives you fifty reference points, dense landmark estimation gives you a continuous map of the body's surface.
Second — and this is the move that matters most for partnered movement — the system explicitly predicts contact probabilities: for each point on each body, the likelihood that it is currently in contact with another body. Contact is no longer a failure mode that confuses the system. It is a signal the system is trained to detect and report.
The training data problem (where do you get thousands of examples of closely interacting bodies with perfect ground-truth annotation?) was solved synthetically: the researchers built a large-scale dataset of simulated multi-person interactions, with exact ground truth available by construction.
Who Gets to Study Movement Now
The deeper significance of markerless capture is not technical but social: it changes who can afford to study movement rigorously.
Marker-based systems cost tens to hundreds of thousands of dollars, require dedicated studio space, and need trained technicians. This concentrated movement research in well-funded labs — primarily in biomechanics, sports science, and the entertainment industry. A dance researcher, an independent somatic practitioner, a small university programme in movement studies: these could rarely access capture at research grade.
A markerless system that works from multi-view video changes the entry cost to a set of consumer cameras and a computer. The capture can happen in the practice space rather than the lab — in the studio where the practice actually lives, without suits, without altering the movement being studied. Practices that were never documented at scientific accuracy because their communities lacked mocap budgets — social dance forms, somatic education lineages, contact-based practices, movement traditions outside the institutional mainstream — become documentable.
There is a historical pattern here worth noticing. Whenever the cost of recording a medium collapses, the diversity of what gets recorded explodes. Cheap audio recording captured musical traditions that formal studios never would have. Cheap video did the same for performance. Markerless capture is positioned to do this for movement — to broaden the archive of what human movement is, beyond what well-funded labs chose to point their cameras at.
What the Cameras Still Cannot See
A necessary caution, familiar to readers of this series: capturing the positions of bodies — even hundreds of dense landmarks per body, even through contact — is not the same as capturing the experience of movement.
The new systems record where bodies are with unprecedented accessibility and accuracy. They do not record the weight exchanged through a point of contact, the muscular effort beneath a visually quiet posture, or the felt anticipation before a movement begins. Those signals live below the skin, in dimensions cameras cannot reach — accessible, partially, to wearable sensors like EMG, and fully only to the moving body itself.
The right way to understand this week's development is as one layer of a fuller picture: the outside view of movement is becoming radically cheaper and more accurate. The inside view — the one somatic practice cultivates and the one movement AI still mostly lacks — remains the frontier.
Further Reading
Velasquez, H. C., Yiannakidis, A., Shin, S., et al. (2026). MAMMA: Markerless accurate multi-person motion acquisition. In Proceedings of CVPR 2026. https://mamma.is.tue.mpg.de/
Zhang, C., et al. (2026). Efficiently reconstructing dynamic scenes one D4RT at a time. In Proceedings of CVPR 2026. https://d4rt-paper.github.io/