Robotics
Elevation map
Definition
An elevation map represents terrain as surface height, and often height uncertainty, over cells in a horizontal grid. Robots use it to reason about slopes, steps, footholds, and clearance, but a single height per cell cannot represent every three-dimensional structure.
Also known as: Terrain elevation map, 2.5D elevation map
Updated
Project terrain into a height grid
An elevation map divides the ground plane into cells and stores an estimated surface height in each cell. A probabilistic map can also store variance or other confidence information. Depth cameras, stereo vision, or lidar provide a point cloud, while the robot pose places each measurement into the map.
Fankhauser and colleagues describe probabilistic terrain mapping that accounts for uncertain robot localisation. The accompanying ANYbotics implementation maintains a local robot-centred map, fuses distance measurements, and tracks height variance as pose uncertainty accumulates.
Support locomotion and local navigation
A wheeled robot can score steepness, roughness, or step height from neighbouring cells. A legged robot can use the surface estimate for footstep planning, body-clearance checks, and terrain-aware control. Keeping the grid centred on the robot bounds memory and preserves local resolution as the robot moves.
An elevation map differs from an occupancy grid, which records whether space is occupied, and from the raw point cloud that supplied measurements. It is a compact surface model. Traversability, foothold quality, and semantic terrain class are derived judgements, not facts guaranteed by the height layer itself.
A single surface loses vertical structure
A conventional 2.5D cell cannot simultaneously store the road, the top of a table, and the underside of a bridge at the same horizontal location. Overhangs, caves, vegetation, thin obstacles, and vertical walls therefore need another representation or extra layers.
Occlusion leaves unknown cells, and pose drift can smear sharp terrain features. Resolution trades geometric detail against compute and map size. Sensor bias, reflective surfaces, moving objects, and delayed transforms can all corrupt the height estimate, so locomotion systems need uncertainty handling and conservative behaviour when the map is incomplete.
Sources
Related terms
Occupancy grid
An occupancy grid divides space into cells and records occupancy information for each cell. A two-dimensional robot map commonly distinguishes occupied, free, and unknown regions.
Point cloud
A point cloud is a collection of points representing sampled locations in space, usually with three-dimensional coordinates. Individual points may also carry attributes such as color or return intensity.
Footstep planning
Footstep planning selects a sequence of foot poses and a contact order that lets a legged robot move towards a goal. A planner must respect reachability, collision, terrain support, and the assumptions of the downstream balance controller.
Lidar
Lidar measures distance using emitted laser light and its return from surfaces. Repeated range measurements across directions can form a spatial scan or three-dimensional point cloud.
State estimation
State estimation infers quantities describing a robot or its environment from measurements and a model. A robot state may include position, orientation, velocity, and other variables that are not all directly measured.