← Back to creations

Robotics & simulation

SLAM & Kalman Localization

A state estimate becomes more useful when it knows what it does not know.

Estimate, observe, correct.

Ready to inspect
Start the filter, then steer the path.
— Ground truth— Kalman estimate· Noisy observations95% covariance ellipse

Observation model

Turn observations off to see uncertainty grow while the filter predicts. Drag in the map or use its arrow keys to steer.

Measured error

Filtered RMS
Observation RMS

Map units. This study isolates 2D localization with a constant-velocity Kalman filter; it is not a complete SLAM implementation.