Camera-based movement intelligence

Physio feedbackbuilt for your body.

Pausen rebuilds your movement in 3D from a short exercise video and turns it into clear, evidence-backed corrections. No wearables, no markers, no lab.

Explore the assessmentSee how it works
09movement cases
03synced views
01clear verdict
Pausen live00:03.6
Overall verdictCorrect & controlled
Scroll to inspect
01 / LIVE DEMO

One movement.
Three synced views.

Watch the source video, the tracked joints, and the 3D reconstruction on one shared timeline. Then read the feedback Pausen delivers.

Choose a case
Correct Refine Needs correction
CASE 02Seated lateral neck stretch
SourceRGB
Landmarks2D pose
Structure3D body
00:00.000:04.2 Synced

In production, the AI receives compact extracted measurements, not a full 3D stream. The 3D view is shown here to make the evidence easy to see.

COMPACT EVIDENCE PACKET

What the AI critic receives

A task-specific packet of validated measurements, representative timepoints, and confidence limits, paired with the source clip and the exercise protocol.

Case
02
Confidence
High
Trunk sway
0.2392
Pose jumps
0
In-frame
100%
Packet size
~20 KB
Open full JSON packet 
{
  "schema_version": "1.0",
  "movement": "lateral_neck_stretch",
  "segment_summary": {
    "stability": {
      "trunk_sway": 0.2392,
      "abrupt_pose_steps": 0
    },
    "quality": {
      "keypoints_in_frame": 1,
      "confidence_limitation": "No per-joint detector confidence in export"
    },
    "extracted_trace": {
      "name": "Head–pelvis offset",
      "representative_values": [
        0.653,
        0.928,
        1.309,
        1.327,
        1.093,
        1.078,
        1.079,
        1.093
      ]
    }
  },
  "critic": {
    "evidence_sufficient": true,
    "interpretation": "Needs correction",
    "known_limitations": [
      "Short clip",
      "Camera-only assessment"
    ]
  }
}
02 / HOW IT WORKS

Measure what matters.
Look closer only when needed.

Pausen runs a layered pipeline. It starts with lightweight pose estimation, validates every measurement, and escalates to full 3D reconstruction only when it can change the clinical answer.

01

Know the exercise

Each movement maps to a clinical protocol: phase structure, target joint ranges, and the compensation patterns the system must rule out.

Protocol routing · movement phase model
02

Track and measure

Monocular video becomes per-joint kinematic traces. Measurements pass temporal-smoothness and anatomical-plausibility checks before they count.

2D/3D pose estimation · confidence gating
03

The AI critic decides

A vision-language model audits the validated evidence and asks one question: is this sufficient for a verdict? Full 3D body reconstruction runs only when it can change the criterion.

VLM evidence critic · adaptive escalation
04

Explain the fix

Scoring stays rule-grounded and auditable. The output is an overall verdict, prioritized corrections in patient language, and a clinician-readable evidence trail.

Deterministic scoring · clinical scorecard

The critic works from a compact evidence packet of roughly 20 KB per movement segment. Full pipeline internals are shared in partner conversations.

THE LAST MILE

Not more motion data.
Better next movements.

Pausen connects objective camera evidence to the one thing patients need most: a correction they can understand and act on.

Replay the demo