The Chicken-and-Egg Problem

Imagine being blindfolded and dropped into an unfamiliar room. To make a map, you first need to know where you are. But to know where you are, you need a map. This is the central problem of mobile robotics — and the answer is SLAM, or Simultaneous Localisation and Mapping.

SLAM is what lets your Roborock S8 MaxV Ultra map your home on its first run, what allows a delivery robot to navigate a sidewalk, and what makes self-driving cars possible. Despite the intimidating name, the underlying idea is surprisingly intuitive.

The Core Loop

Every SLAM system, regardless of which sensors it uses, runs the same loop several times per second:

  1. Observe — Take a sensor reading of the surroundings.
  2. Predict — Estimate where the robot has moved since the last reading (using wheel encoders, gyroscopes, or visual cues).
  3. Match — Compare what the sensors see now with what they have seen before. If the robot recognises a feature it has mapped earlier, it knows where it is on that map.
  4. Update — Adjust both the robot’s position estimate AND the map, using the new observation to make both more accurate.

That fourth step is the magic. Each new observation corrects errors in both the robot’s self-position AND its understanding of the world.

Two Main Flavours

Visual SLAM (vSLAM)

Visual SLAM uses one or more cameras to identify distinctive visual features — corners of furniture, doorways, picture frames — and track how they move between frames. As the robot moves, it triangulates the position of those features in 3D space.

vSLAM is cheap (cameras are commodity hardware) but struggles in low light, blank walls, and rapidly changing scenes. Most consumer robots use a hybrid approach to compensate.

LiDAR SLAM

LiDAR SLAM uses a spinning laser scanner to measure exact distances to surrounding surfaces. Each scan produces a point cloud — thousands of distance measurements forming a 2D or 3D outline of the room. SLAM algorithms match new point clouds against the existing map to localise the robot precisely. We go deeper on the differences in our companion piece: LiDAR vs Camera Navigation.

LiDAR is faster and more accurate than vision, especially in cluttered or dim environments. Premium robot vacuums like the Dreame L30 Ultra and the Narwal Freo X Ultra rely on LiDAR for centimetre-accurate mapping.

Why “Loop Closure” Matters

The hardest part of SLAM is recognising when you have returned to a place you have been before. This is called loop closure. Without it, small drift errors compound — your robot thinks the hallway is 8 metres long the first time and 9 metres the second time, and the map slowly distorts.

When SLAM detects a loop closure, it goes back and “stitches” the map together, redistributing the accumulated error along the path. Suddenly, the map snaps into a consistent, accurate shape.

Where SLAM Fails

Even the best SLAM has weaknesses. Long, featureless corridors give few landmarks to track. Glass walls confuse both lasers and cameras. Cluttered environments with constantly moving objects (people, pets, vehicles) can degrade map quality.

The next generation of robots increasingly fuses SLAM with semantic understanding — knowing that the object moving across the room is a person rather than a permanent feature — so the map remains stable even as the world changes.

The Bottom Line

SLAM is the foundational technology that turned mobile robots from research curiosities into mass-market products. Every time your robot vacuum efficiently covers your floor in neat rows, every time a delivery drone returns precisely to its launch pad — SLAM is the reason. It is one of the great engineering achievements of the last two decades, and the techniques are still evolving rapidly.