Map Merging

name: map-merging description: 'Merge multiple maps. Use when combining maps from different sessions, robots, or partial mapping runs.'

Map Merging Techniques

Overview

Map merging combines maps from multiple sessions, robots, or partial scans into a single coherent map. This is necessary when a single mapping run cannot cover the entire environment, or when multiple robots map simultaneously. There is no one-size-fits-all solution—the approach depends on whether you have pose graph data, overlapping regions, or only raw images.

Approach 1: SLAM Toolbox Deserialization + Continue Mapping

The most robust approach when using SLAM Toolbox. You serialize a partial map, then start a new session that loads the serialized graph and continues mapping.

Workflow

# Session 1: Map area A
ros2 launch slam_toolbox online_sync_launch.py params_file:=params.yaml
# Drive through area A, then serialize:
ros2 service call /slam_toolbox/serialize_map slam_toolbox/srv/SerializePoseGraph \
  "{filename: '/home/robot/maps/area_a'}"

# Session 2: Continue into area B
# Configure SLAM Toolbox to load the serialized graph:

slam_toolbox:
  ros__parameters:
    map_file_name: /home/robot/maps/area_a    # Base name, no extension.
    map_start_pose: [0.0, 0.0, 0.0]
    map_start_at_dock: true
    mode: mapping

# Start SLAM Toolbox with the above config.
# Drive through area B. The graph now contains A + B.
# Serialize the combined graph:
ros2 service call /slam_toolbox/serialize_map slam_toolbox/srv/SerializePoseGraph \
  "{filename: '/home/robot/maps/area_ab'}"

Requirements

• The robot must start session 2 at a location within the area covered by session 1's map.
• SLAM Toolbox must successfully match the initial scans against the loaded graph.
• If the robot starts outside the mapped area, merging fails silently (no constraints connect old and new nodes).

Verifying the Merge

After deserialization + continued mapping, check the constraint visualization in RViz. There should be constraints connecting old (area A) nodes to new (area B) nodes at the overlap region.

Approach 2: multirobot_map_merge Package

The multirobot_map_merge package merges OccupancyGrid maps from multiple sources using feature matching on the map images.

sudo apt install ros-jazzy-multirobot-map-merge   # If available for Jazzy.

Configuration

map_merge:
  ros__parameters:
    robot_map_topic: map
    robot_namespace: ''
    merged_map_topic: /merged_map
    world_frame: map
    known_init_poses: true        # true if you know the relative poses of the robots/sessions.
    init_pose_x_0: 0.0
    init_pose_y_0: 0.0
    init_pose_yaw_0: 0.0
    init_pose_x_1: 10.0           # Robot 2 starts 10m east of robot 1.
    init_pose_y_1: 0.0
    init_pose_yaw_1: 0.0
    merging_rate: 0.5             # Hz. How often to attempt merging.
    estimation_rate: 0.5          # Hz for transform estimation (if known_init_poses=false).
    estimation_confidence: 1.0    # Minimum confidence for automatic alignment.

Known vs Unknown Initial Poses

• known_init_poses: true: You provide the relative starting positions of each robot/session. The package directly overlays the maps using these transforms. More reliable.
• known_init_poses: false: The package attempts to find the alignment automatically using feature matching on the occupancy grids. Works if there is sufficient overlap (~30%) and distinctive features. Often fails in symmetric or featureless environments.

Limitations

• Only merges 2D OccupancyGrids—no pose graph, so no correction of internal drift.
• Requires significant overlap between maps for automatic alignment.
• Quality depends on individual map quality. Distorted maps produce distorted merges.

Approach 3: Manual Image Merging

For simple cases, manually aligning map images in an editor can be effective.

Steps

1. Open both PGM/PNG maps in GIMP or similar.
2. Identify overlapping features (walls, doorways).
3. Align the second map over the first using translate/rotate.
4. Merge layers: for each pixel, take the most informative value (occupied > free > unknown).
5. Export as PGM.
6. Manually create the YAML metadata file with the correct origin and resolution.

Resolution Matching

Both maps must have the same resolution. If they differ, rescale one:

# In GIMP: Image → Scale Image → set width/height to match the resolution ratio.
# Or use ImageMagick:
convert map_b.pgm -resize 200% map_b_scaled.pgm   # If map_b is 0.1m and map_a is 0.05m.

Origin Alignment

After merging, the origin must be recalculated. The new origin is the bottom-left corner of the merged image in map coordinates:

# If map_a origin is (-12.2, -10.7) and map_b was shifted by (15.0, 0.0):
# New origin depends on which map's bottom-left is further negative.
new_origin_x = min(origin_a_x, origin_b_x + shift_x)
new_origin_y = min(origin_a_y, origin_b_y + shift_y)

Approach 4: Cartographer Map Merging

Cartographer supports multi-trajectory mapping natively. You can add a second trajectory that shares the same pose graph:

# Start trajectory 1 for robot 1:
ros2 service call /start_trajectory cartographer_ros_msgs/srv/StartTrajectory ...

# Start trajectory 2 for robot 2 (or second session):
ros2 service call /start_trajectory cartographer_ros_msgs/srv/StartTrajectory ...

Both trajectories contribute to the same global optimization. The constraint builder finds inter-trajectory loop closures where the trajectories overlap.

Overlapping Region Requirements

For any automated merging to work:

Map Georeferencing

For outdoor or multi-building scenarios, maps can be aligned to a global coordinate system (GPS/UTM):

1. Record GPS coordinates at known points during mapping.
2. Compute the affine transform from map coordinates to UTM.
3. Store the transform in the map metadata or a separate config.

This enables merging maps from different buildings or areas by aligning them in the shared global frame.

Best Practices

• Always serialize the full SLAM graph, not just the image. Images lose internal pose graph structure.
• Test merges incrementally: merge two maps, verify, then add a third.
• Prefer graph-based merging (SLAM Toolbox deserialization, Cartographer multi-trajectory) over image-based when possible.
• Validate with loop closures: after merging, drive the robot through the overlap region to confirm the merge is geometrically correct.
• Version your maps: keep dated copies of serialized graphs so you can revert a bad merge.