Robotics
Loop closure
ActivePublished
Key
innovation
Turning place re-recognition into a global constraint that corrects drift across the entire map.
Category
Robotics
Abstraction level
Building block
Operation level
Robot control
How it works
The system recognizes that the current observation matches a previously mapped place, geometrically verifies the match and adds an edge to the pose graph. Graph optimization distributes the error along the whole loop. False closures (e.g. between identical-looking places) are dangerous and require verification.
Problem solved
Odometry and incremental estimation inevitably accumulate error (drift), causing long trajectories and maps to diverge. Loop closure provides a strong global constraint that 'stitches' the map together.
Key mechanisms
Place recognition (appearance/geometry)
Geometric verification of the match
Adding an edge to the pose graph
Global graph optimization