How VR Full-Body Tracking Works

Drag the mannequin to orbit. Toggle trackers to see which joints are truly tracked, which are solved by inverse kinematics, and which are pure guesses. Then crank the latency slider and watch the avatar fall behind its own trackers.

3 tracked points · fidelity 45%
drag to orbit · scroll or pinch to zoom

Tracker Loadout

Headset (HMD)
Hand controllers
Hip tracker
Foot trackers (pair)
Chest tracker
Knee trackers (pair)
Elbow trackers (pair)
Show base station sweeps
Pose fidelity
Estimated by IK: elbows. Guessed: hips, knees, feet.
trackedIK-solvedguessed

Lighthouse tracking, step by step

  1. Sync flash. A base station floods the room with an IR flash to mark time zero.
  2. Laser sweep. A spinning rotor sweeps a laser line across the room, horizontally then vertically, about 60 times per second.
  3. Timing. Each photodiode on a tracker records when the beam hit it. Time since flash converts directly to an angle: angle = t × rotor speed.
  4. Pose solve. With 20+ diodes at known positions on the tracker shell, two angles per diode are enough to solve full 6-DoF position and rotation to sub-millimetre precision.
  5. IMU fusion. Between sweeps, a 1000 Hz gyro + accelerometer predicts motion; sweeps correct the drift.

Inverse kinematics fills the gaps

SteamVR only knows where the trackers are. Everything between them is solved by IK: given the position of a hand and shoulder, where must the elbow be?

  • A chain of 2 bones with fixed lengths has a circle of valid elbow/knee positions. IK pickers use heuristics (elbows point down-and-out, knees point forward).
  • 3-point tracking (head + hands): hips and legs are fully guessed from head height, so avatar legs drift and slide.
  • Adding a hip tracker anchors the spine; foot trackers make legs a solvable 2-bone chain (only knees remain heuristic).
  • Full 10-11 point setups leave almost nothing to guess, which is why dancers and mocap users wear them.

The latency budget

Total motion-to-photon delay is a pipeline. Roughly, for a lighthouse system:

StageTypical cost
IMU sample + fusion1–2 ms
Radio to PC (2.4 GHz)2–7 ms
IK solve + game frame5–11 ms
Render + display scanout8–12 ms

Under ~20 ms feels instant. Past ~50 ms your own body feels wrong; past 100 ms full-body avatars visibly swim, which is exactly what the latency slider above simulates.

Tracking tech compared

MethodHowTrade-off
Lighthouse (Vive/Tundra)Swept lasers + photodiodesBest precision; needs base stations
Inside-out cameraHeadset cameras watch the room and bodyNo setup; loses limbs it cannot see (occlusion)
IMU-only (SlimeVR-style)Gyro + accel per limb, no external refCheap, occlusion-free; yaw drift needs periodic reset

Why avatars glitch: occlusion and drift

  • Occlusion. Lighthouse trackers need line-of-sight to at least one base station. Crouch behind a couch and your foot tracker goes blind; the IMU coasts for a moment, then the pose snaps when a sweep is reacquired.
  • Reflections. Mirrors and glass bounce the laser sweep, producing ghost hits. That is why setup guides say to cover mirrors.
  • IMU drift. Gyroscopes integrate angular velocity, so tiny bias errors accumulate into yaw drift. Lighthouse corrects it every ~8–16 ms; IMU-only suits must periodically face forward and reset.
  • Calibration slop. Trackers strap to clothing, and clothing moves. A hip tracker that rotates 10 degrees on a waistband makes the whole avatar lean. Good software lets friends or the wearer re-calibrate with a T-pose in seconds.

Quick self-check

1. With only headset + controllers, how does the software place your hips?

2. Why does a 2-bone IK chain (hip to foot) still need a heuristic?

3. Which latency figure is generally considered imperceptible?

Glossary

  • 6-DoF — six degrees of freedom: position (x, y, z) plus rotation (pitch, yaw, roll).
  • Pose — the complete set of joint positions and rotations for one frame.
  • FBT — full-body tracking; commonly 3-point (no FBT), 6-point (hip + feet), or 10–11 point setups.
  • IK / FK — inverse kinematics works backward from an end effector to joint angles; forward kinematics goes the other way.
  • Motion-to-photon — time from your body moving to the display showing that movement.
  • End effector — the tracked point an IK chain must reach, e.g. your hand or foot.
  • Sync flash / sweep — the lighthouse timing reference and the rotating laser plane that follows it.

From mocap stages to living rooms

  • Optical mocap (film/games): dozens of ceiling cameras track retroreflective dots at sub-millimetre accuracy. Superb, but a studio costs more than a house.
  • 2016, Lighthouse 1.0: Valve inverted the problem - instead of cameras watching markers, cheap sensors on the body watch structured light from two small emitters. Room-scale 6-DoF for consumers.
  • 2018+, Vive Trackers: the same pucks that tracked controllers strap to hips and feet, and social VR communities turned full-body avatars into a culture of dance, mirrors, and motion-captured performance.
  • 2022+, IMU suits: open-source projects proved that gyro-only trackers plus clever skeletal constraints get 90% of the experience with zero base stations.
  • Now: headset cameras plus ML body estimation infer legs with no hardware at all - the guessed joints in this demo are increasingly guessed by a neural network instead of simple heuristics.

Common setups and what they cost you

SetupPointsWhat is guessed
Headset only1Everything below the neck
+ Controllers3Elbows (IK), whole lower body
+ Hip4Legs entirely, elbows (IK)
+ Feet6Knees and elbows via IK only
+ Chest, elbows, knees10–11Almost nothing

Try reproducing each row with the toggles above and watch the fidelity meter agree.

Enjoy this tool? Build your own with Super