Concept
SLAM
Simultaneous Localisation and Mapping - the robotics problem of building a map of an unknown environment and determining the robot's position within it at the same time, without any external reference like GPS.
Added May 18, 2026
A fundamental challenge for any autonomous robot is knowing where it is. GPS provides global positioning outdoors but is unavailable indoors, unreliable in dense urban canyons, and too imprecise for close-quarters manoeuvring. SLAM (Simultaneous Localisation and Mapping) is the computational problem of answering two interdependent questions simultaneously: "What does the environment look like?" (mapping) and "Where am I within it?" (localisation).
The chicken-and-egg nature of SLAM makes it technically challenging. To accurately determine your position in a map, you need an accurate map. To build an accurate map, you need to know your position accurately. SLAM algorithms must jointly estimate both from raw sensor data - typically laser range finders (LiDAR), cameras, wheel odometry, and inertial measurement units (IMUs).
Feature-based SLAM extracts distinctive features (corners, edges, visual landmarks) from sensor observations, matches them across time to track motion, and maintains a probabilistic estimate of robot pose and feature positions. The Extended Kalman Filter (EKF-SLAM) and its variants maintain a joint Gaussian distribution over robot pose and landmark positions, updating it as new measurements arrive. The key challenge is data association: determining which current observation corresponds to which previously seen landmark.
Graph-based SLAM (pose graph optimisation) represents the problem as a graph where nodes are robot poses at different times and edges are spatial constraints from sensor measurements. Loop closure detection - recognising when the robot has returned to a previously visited location - adds constraints that reduce accumulated drift and correct the map. Optimisation methods (like g2o or GTSAM) then find the set of poses that best satisfies all constraints.
Visual SLAM (V-SLAM) uses cameras as the primary sensor. ORB-SLAM2/3 are widely used monocular/stereo/RGB-D visual SLAM systems. LiDAR-based SLAM (like Cartographer from Google, or LeGO-LOAM) uses 3D point clouds for more robust metric maps.
Deep learning approaches to SLAM learn feature extraction and correspondence matching from data, often outperforming hand-engineered features in challenging conditions (darkness, texture-poor environments). Neural SLAM approaches learn end-to-end mappings from observations to maps and pose estimates.
Applications: autonomous vehicles (HD map building and real-time localisation), mobile robots (indoor navigation for delivery robots, warehouse AMRs), augmented reality (anchoring virtual content to real-world geometry), and drone navigation.
Analogy
Exploring an unfamiliar building with no map, in the dark, while keeping a hand-drawn map of where you have been. As you move forward, you add to your map (mapping). When you reach a junction, you estimate your position based on your map and the distances you have walked (localisation). If you recognise a distinctive junction you visited before, you can correct any accumulated drift in your estimated position (loop closure). You are simultaneously building the map and using it to locate yourself - SLAM in action.
Real-world example
A warehouse autonomous mobile robot (AMR) starts each shift by performing SLAM as it drives its initial path. LiDAR scans build a 2D floor plan of the warehouse layout, with shelf positions, obstacles, and open aisles as landmarks. Odometry from wheel encoders estimates motion between scans. When the robot completes its first circuit and returns to its starting position, loop closure detects the return and corrects any drift accumulated over the circuit. The resulting map is saved and used for all subsequent navigation during the shift.
Why it matters
SLAM is one of the foundational capabilities that makes autonomous robotics possible. Without SLAM, a robot cannot navigate an unstructured environment or maintain situational awareness of its position relative to its surroundings. Understanding SLAM explains why autonomous vehicles need extensive mapping infrastructure, why indoor robot navigation requires different approaches from outdoor navigation with GPS, and why loop closure and data association are central research problems in robotics.
In the news
No recent coverage - check back later.
Related concepts