READY · Next: prediction
t = 0.0 s · cycle 0
Drag to

Drag / right-drag to orbit · Shift-drag to pan · Scroll to zoom

Touch: one finger orbits · two fingers zoom and pan

Visual legend
True state & path
Noisy GPS measurement
Kalman state estimate
Predicted state · translucent marker
Red disk · Mahalanobis d ≤ 2
Orange disk · d ≤ 4
Yellow disk · d ≤ 6

The translucent marker shows the prediction before GPS correction; the regions share its center. Track mode follows straights, sweeping turns and a tighter infield section.

Noise & playback

dt = 0.1 s per cycle. At 1×, each phase takes 0.25 s for inspection. Q tunes the filter; R applies to the next GPS sample.

The cycle: predict motion → grow uncertainty → receive GPS → calculate gain → correct the state and covariance.

Edit process covariance Q

These are the actual variances added per prediction, at the current green slider scale. Each x/y pair can have different uncertainty and correlation (−1 to +1). The filter requires a symmetric, positive semidefinite matrix.

State pairVariance xVariance yCorrelation ρ

Off-diagonal covariance is ρ√(variance x × variance y). Editing Q changes uncertainty, not the physical track. The existing P evolves from the next prediction; Q is not the same matrix as P.