SLAM: Building Maps in Real Time

What SLAM actually solves

SLAM stands for Simultaneous Localization and Mapping — the problem of building a map of an unknown environment while at the same time tracking your position within it. The tricky part: you need the map to locate yourself, but you need to know your location to build the map accurately. SLAM algorithms solve this chicken-and-egg problem probabilistically.

For mobile robots, the most practical SLAM approach is 2D LIDAR-based: a spinning LIDAR (RPLIDAR A1, YDLIDAR X4) produces a 360° scan every 100-200ms, and slam_toolbox matches each new scan against previous ones to estimate motion and build a 2D occupancy grid.

The occupancy grid

slam_toolbox outputs a nav_msgs/OccupancyGrid message on the /map topic. Each cell in the grid stores a value: 0 = free space, 100 = obstacle, -1 = unknown. At 5cm resolution (the slam_toolbox default), a typical room produces a grid of ~2000×2000 cells.

Tuning slam_toolbox for small spaces

Default slam_toolbox params are tuned for large open spaces. For home or office environments (room sizes 3-10m), adjust these:

slam_toolbox:
  ros__parameters:
    # Resolution: 5cm for indoor, 10cm for large spaces
    resolution: 0.05

    # How far the robot must move before processing a new scan
    # Smaller = more frequent updates, higher CPU
    minimum_travel_distance: 0.05   # meters (default 0.5 — too coarse)
    minimum_travel_heading: 0.05    # radians

    # Loop closure — how aggressively to correct accumulated drift
    # Higher = better maps, more CPU
    link_match_minimum_response_fine: 0.45

    # Scan buffer: hold more scans for better matching
    scan_buffer_size: 20
    scan_buffer_maximum_scan_distance: 5.5  # match RPLIDAR A1 range

    # Pose covariance: trust odometry more if encoders are good
    transform_timeout: 0.2
    tf_buffer_duration: 30.0

The SLAM → nav2 handoff

SLAM builds the map. Once you have a good map, you switch to nav2 + AMCL (Adaptive Monte Carlo Localization) for autonomous navigation. The flow:

Minimal nav2 launch with saved map

# After saving map to ~/maps/room.yaml + room.pgm
ros2 launch nav2_bringup bringup_launch.py \
  map:=/home/robot/maps/room.yaml \
  use_sim_time:=false \
  params_file:=./nav2_params.yaml

# In RViz2: set initial pose with "2D Pose Estimate" tool,
# then send navigation goals with "Nav2 Goal" tool