What is SLAM?

A chicken-and-egg problem

To put a wall on a map, a robot needs to know where it was when it saw the wall. To know where it is, it needs a map. SLAM, short for simultaneous localisation and mapping, does both at once. Durrant-Whyte and Bailey’s 2006 tutorial already called the general method a well understood and established part of robotics, while noting practical issues that remained, especially in complex outdoor places.

How it works, step by step

1. Estimate the motion. Wheel turns, a gyroscope and accelerometer, or the way the floor slides past a camera give a first estimate of how far the robot moved: what optical flow is.
2. Look around. A camera, a laser scanner or time-of-flight sensors measure walls, corners and furniture.
3. Match. The robot compares what it sees with the map so far: is this the corner it saw a minute ago?
4. Correct both. Where the motion estimate and the measurements disagree, it adjusts its position and the map together, keeping track of how unsure it is about each.

Drift, and closing the loop

Every motion estimate is a little wrong, and the errors add up: after a long trip round a flat, a robot that only counted its wheel turns would draw the kitchen in the wrong place. The fix comes when it recognises a place it has already mapped. That “loop closure” tells it how far it has drifted, and the correction can be spread back over the whole path, straightening the map behind it.

Where you meet it

• Robot vacuums that show a floor plan in their app and clean room by room.
• Augmented reality on phones and headsets, which pins a virtual sofa to your real floor as you walk round it.
• Drones and robots indoors, where satellite positioning does not reach.
• Self-driving cars, which combine maps made in advance with what they see live.

What is still hard

In a 2016 survey, Cadena and colleagues described thirty years of striking progress and a steady move into industry, and still set out open problems, among them robustness in long-term mapping. They also asked whether SLAM is solved; their answer, roughly, is that it depends on the robot, the place and how well it has to work. In a home, the hard parts are familiar:

• Things move. Chairs, doors, bags and people change the scene from one trip to the next.
• Places look alike. Two plain white corridors are easy to confuse.
• Glass and mirrors fool light-based sensors: why glass and mirrors confuse robots.
• Light changes from day to night, which matters for cameras.

A map of your home is personal

A floor plan shows how big your home is, how it is laid out and, over time, where people spend their day. Before you bring a mapping robot home, ask where the map is kept, whether it leaves your home, whether you can delete it, and who else can see it: home robots and camera privacy.

How Alkyon is designed

Alkyon, our concept for a flying home companion, is designed to build a 3D map of your home from its own depth sensors and to keep that map on its dock, in your home. In our simulation it plans its routes on that map and does not fly towards space it has not seen; correcting its position against the map is one of our open tasks. It is a concept in development, for Europe only: how indoor drones navigate.

Sources, read 4 November 2026: Durrant-Whyte and Bailey, simultaneous localization and mapping: part I, IEEE Robotics & Automation Magazine, 2006; Cadena and colleagues, past, present, and future of simultaneous localization and mapping, IEEE Transactions on Robotics, 2016. General information.