slam-toolbox-online

Use the slam_toolbox::OnlineSyncSlamToolbox plugin and set odom_frame, map_frame, base_frame, and scan_topic to match your robot. The node publishes the map to odom transform while your robot must already publish odom to base_link.

OnlineSyncSlamToolbox processes each laser scan as it arrives for real-time mapping. OnlineAsyncSlamToolbox queues scans and processes them in a background thread, which helps when the CPU cannot keep up but the map may lag behind reality.

Adjust loop_search_space_dimension, loop_match_minimum_response_coarse (0.35), and loop_match_minimum_response_fine (0.45) to control detection sensitivity. A successful loop closure triggers full graph optimization that redistributes accumulated drift.

The Ceres solver can use deep recursion on large pose graphs, exhausting the default stack. Increase stack_size_to_use to 40000000 (40MB) or more, and use SPARSE_NORMAL_CHOLESKY as the linear solver for maps with over 1000 nodes.

Serialize the graph in interactive mode, then restart with mode set to localization and map_file_name pointing to the saved map without extension. The node then scan-matches against the loaded graph without adding new nodes.

Increase minimum_time_interval between processed scans and raise minimum_travel_distance and minimum_travel_heading so scans are only added after meaningful motion. These thresholds also prevent graph bloat when the robot is stationary.