SLAM
Estimating a map and a trajectory at once, and the structure that makes it tractable.
SLAM is not a single algorithm. It is a problem — estimate where you are while building the map you are estimating against — with a characteristic structure that every solution exploits: the measurements are local, so the information matrix is sparse, and the problem has unobservable directions that must be handled deliberately rather than ignored.
This topic is about that structure. The front end (features, matching, tracking) is a large subject in its own right; here the interest is in what the back end is actually solving and why it is shaped the way it is.
- 01 What SLAM is estimating Core 9′ A trajectory and a map, jointly, from measurements that only ever relate a few variables at a time. Repairs Thinking of SLAM as "localisation plus mapping run alternately" rather than as one joint estimate whose correlations are the whole difficulty.
- 02 Factor graphs and sparsity Core 1 interactive 11′ Draw the graph and you have drawn the matrix. Eliminate a variable and you can see what it costs. Repairs Treating the factor graph as a diagram for papers, rather than as a literal picture of the linear algebra.
- 03 Bundle adjustment Advanced 10′ The reprojection error is the residual, and everything else is bookkeeping about which Jacobian block goes where. Repairs Regarding bundle adjustment as a mysterious refinement step rather than as the same least-squares problem with an obvious residual.