Topological Map

• Also called: topological graph, topometric map, place graph
• Category: State estimation & navigation › Navigation
• Tags: navigation, mapping, graph-search, visual-navigation, mobile-robots

A map that represents an environment as a graph of distinct places connected by traversable links, rather than as precise geometry, so a robot can plan routes between them.

Definition

Topological maps represent an environment as a graph. Nodes stand for distinct places, such as rooms, doorways, intersections or remembered viewpoints, and edges record that the robot can travel directly between two places. Unlike a metric map, such as an occupancy grid, a topological map does not need globally consistent coordinates; it captures how places are connected, not exactly where they are.

This mirrors how people often describe routes: go down the corridor, turn at the kitchen, the lab is the second door. Benjamin Kuipers' Spatial Semantic Hierarchy is a well-known framework that layers control, causal, topological and metric knowledge in this spirit. Many practical systems use hybrid topometric maps, keeping a graph for long-range planning and local metric maps for precise motion within each node.

Navigation with a topological map has two levels. A graph search, such as Dijkstra's algorithm or A*, finds a sequence of nodes from the current place to the goal. A local controller then moves the robot from one node to the next, often by recognizing the target place from sensor data. In learning-based visual navigation, nodes frequently store camera images, and a learned policy drives toward the image of the next node.

Topological maps scale well to large spaces and tolerate drift in odometry, because errors do not accumulate across a global frame. Their weaknesses are perceptual aliasing, where two different places look alike, and missing edges, where a traversable connection was never observed. Loop closure, the recognition that the robot has returned to a known place, keeps the graph consistent.

Why it matters

Topological maps let robots navigate large environments with compact memory and without perfect global localization. They are a natural fit for learned visual navigation, where a robot remembers places by what they look like rather than where they are in a coordinate frame.

Key applications

• Long-range indoor and outdoor navigation with image-based place memory
• Route planning over large buildings or campuses
• Exploration, where newly found places are added as nodes
• Hierarchical planning that combines graph routes with local metric control

In practice

A team deploying a visual navigation robot might first drive it through a building while saving camera images every so often as graph nodes, linking consecutive images and adding extra edges where a learned model predicts two images are reachable from each other. At run time the robot localizes itself to the nearest node, searches the graph for a route to the goal image, and uses a learned policy to drive toward each successive node image, re-localizing as it goes.

Common misconceptions

• Topological maps contain no geometry at all — many store relative poses or local metric maps at nodes; what they drop is a single global coordinate frame
• Topological and metric maps are mutually exclusive — hybrid topometric maps are common in practice
• A topological map is just a list of waypoints — it is a graph, so it supports alternative routes and replanning when an edge is blocked

Who should know this

• Mobile robot navigation engineers
• Researchers working on learned visual navigation
• SLAM and mapping engineers

Related technologies

• SLAM
• Occupancy Grid
• Graph Search (A*, Dijkstra)
• Visual Place Recognition
• Loop Closure

Related terms

• SLAM
• Frontier-Based Exploration
• Motion Planning

Related projects

• LingBot-Map: Streaming 3D reconstruction from one camera: pose and depth for every frame, live
• Cartographer: Google's real-time 2D and 3D SLAM system for multi-platform robotics
• OpenDroneMap: Open-source photogrammetry toolkit that turns drone images into maps and 3D models
• Nav2: ROS 2 navigation stack with path planning and obstacle avoidance for mobile robots
• DimOS: Python robotics runtime with agents, SLAM and teleop, no ROS required
• RTAB-Map: Versatile graph-based SLAM supporting RGB-D, stereo, and LiDAR with real-time operation