A robot moving through a space has to answer three questions to be useful. Where am I? What’s around me? And — the hardest one — how do I figure out both at the same time, with no map and no GPS? Those three questions correspond to three different problems with three different algorithm families: localization, mapping, and SLAM (simultaneous localization and mapping).
It took the robotics community thirty years and a remarkable amount of math to figure out that SLAM is well-posed at all. The breakthrough was the realization that localization and mapping aren’t independent — the correlations between landmarks are what makes the joint estimation tractable. From that observation came everything: EKF SLAM in the 1990s, particle-filter SLAM in the 2000s, graph SLAM in the 2010s, and the modern visual SLAM pipelines that ship in every consumer drone and AR headset.
This primer is a tour of all of it.
What it covers
Eight chapters, eight live demos, about thirty minutes to read.
§1 — Three questions, three problems. The setup. Why localization, mapping, and SLAM are different problems with different tools, and the historical reason they weren’t always understood as such.
§2 — Localization (“Where am I?”). Three demos: particle filters (Monte Carlo localization, the workhorse of every mobile robot), Kalman localization with known landmarks, and AMCL (the adaptive variant that ROS ships).
§3 — Mapping (“What’s around me?”). Two demos: occupancy grids (the discrete-cell representation that defines mobile robotics), and feature-based maps with known sensor poses.
§4 — SLAM (“Both at once”). Three demos showing the three eras: EKF SLAM with its quadratic complexity, FastSLAM with particle-per-trajectory, and graph SLAM as the modern factor-graph formulation. The math gets serious here but the figures keep it tangible.
§5 — Loop closure: the trick that saves SLAM. Why drift accumulates, how recognizing a place you’ve been before lets you correct an entire trajectory at once, and the matching algorithms (bag-of-words, learned descriptors) that make it work in practice.
§6 — Side by side. Visual SLAM vs LiDAR SLAM vs visual-inertial. When each one wins. The hardware-and-environment matrix that decides what you should run.
§7 — What a real SLAM stack looks like. ORB-SLAM3, RTAB-Map, the open-source landscape, and the engineering reality behind the academic algorithms.
§8 — References. Reading list — the foundational papers and books worth knowing.
Every chapter has a live demo you can interact with — drag landmarks, move the robot, watch the filter respond.
Read it
The primer is the natural follow-on to the path planning and path tracking posts — those covered deciding where to go and how to follow the path; this one covers how the robot knows where it is in the first place. Together the three of them sketch the full perception-planning-control pipeline of a mobile robot.
← Back to Autonomy