Data Association
When a robot sees a landmark, it has to decide which landmark on its map it is looking at. If landmarks are far apart this is easy. If they are close together, or the robot is unsure where it is, it may match a reading to the wrong landmark, and that mistake can wreck its map or position.
Nearest neighbor
Predict where each map landmark \(j\) should appear, then match each observation \(i\) to the closest one by Mahalanobis distance:
A gate also rejects any match whose distance is too large, leaving that observation unmatched, perhaps a new landmark.
What to notice
With low uncertainty and well-spaced landmarks, every match is right. Raise the uncertainty or crowd the landmarks and wrong matches, red, appear. The gate turns some wrong matches into "unmatched", which is safer than a wrong match. Robust systems use joint compatibility tests, which check that a whole set of matches makes sense together, or unique features like visual descriptors, so landmarks are easier to tell apart.