SLAM MCQ
Track the sensor pose, build a map, close loops when you revisit places, and keep drift under control.
Localize
Pose
Map
Landmarks / grid
Loop closure
Drift fix
Front / back
Tracking + BA
What is SLAM?
Simultaneous Localization and Mapping estimates the sensor trajectory while building a model of the environment. Visual SLAM uses cameras (often with IMU as VIO). Front-end extracts tracks; back-end optimizes poses and landmarks; loop closure detects revisits and corrects accumulated drift.
Drift vs loop closure
Odometry-style updates accumulate error; recognizing a previously seen place adds constraints that globally align the map.
Key ideas
Localization
Where is the camera/sensor in the map frame over time?
Mapping
Sparse landmarks, dense surfels, occupancy grids, or learned implicit maps.
Loop closure
Place recognition + pose-graph or bundle adjustment to reduce drift.
Bundle adjustment
Non-linear least squares over poses and 3D points given observations.
Typical pipeline
Features / direct tracking → keyframes → local BA → loop detection → global optimization