SLAM mini projects.

Global localization for a lost indoor robot in Wean Hall using only odometry and a laser rangefinder against a known map. Monte Carlo localization with an odometry motion model, a beam sensor model, Bresenham ray casting and low-variance resampling, validated across three robot datasets.

Estimating the robot trajectory and six landmark positions simultaneously from range-and-bearing measurements, with an explicit covariance on every one, plus a systematic study of how each noise parameter reshapes the uncertainty ellipses.

SLAM as one large sparse least-squares problem instead of a filter. Benchmarking LU against QR with and without COLAMD reordering, reading the answer off the sparsity of the information matrix, then extending to the nonlinear case.

A full dense RGB-D SLAM system: point-to-plane ICP with projective data association for tracking, point-based fusion into a surfel map at under 9% of the raw point count, then tracking against that map instead of ground truth.