Showcase · Systems
Sensor Fusion Playground
A quaternion error-state Kalman filter (ESKF) fuses a simulated strapdown IMU with GPS, a barometer, a magnetometer, a downward LiDAR altimeter, UWB radio ranging to fixed beacons, and optical flow — rendering the estimate as a quadrotor, the ground truth as a faint ghost, and the filter’s live position-uncertainty ellipsoid, in 3D. Toggle any sensor and drag the sliders: drop GPS and the ellipsoid balloons to metres, then turn on UWB ranging and it snaps back, localised with no GPS at all. The interesting part is doing the quaternion filter right — a minimal body-frame error state, exact finite-difference-checked Jacobians for every sensor, and a covariance that actually matches the error, verified by a Monte-Carlo NEES consistency gate, not just eyeballed.
View source on GitHubDrag to orbit, scroll to zoom. The blue quadrotor is the estimate and the faint green one is ground truth; the orange shell is the 95% position-uncertainty ellipsoid, and the purple markers are the UWB beacons. Toggle sensors in the panel and watch the error hold inside its 3σ envelope on the consistency plot. The native harness (cargo run -p eskf-cli -- check) is what proves the covariance honest.