Skip to main content

Module lidar_graph_slam

Module lidar_graph_slam 

Source
Expand description

2D LiDAR graph SLAM: scan-to-map odometry plus loop closure.

ScanToMapMatcher keeps local drift small but cannot remove it: once the robot leaves the submap radius, errors are frozen into the trajectory. This module adds the global layer on top of it:

  1. Pose-graph nodes are created every node_translation / node_yaw of front-end motion and keep their body-frame scan. Consecutive nodes are linked by odometry edges measured by the front end.
  2. Loop detection: each new node searches older nodes (at least loop_min_node_gap back) within loop_search_radius of its current estimate, and registers its scan against the candidate’s neighborhood map with coarse-to-fine point-to-line ICP.
  3. Verification: a closure is accepted only if the inlier ratio, mean residual, and correction size pass the gates; it then becomes a loop edge and the whole graph is re-optimized with optimize_pose_graph.

The front end keeps running in its own (drifting) frame; only its relative motion between nodes is used, so optimized node poses never disturb the local matcher.

Structs§

LidarGraphSlam
Scan-to-map odometry with pose-graph loop closure.
LidarGraphSlamConfig
Tuning parameters for LidarGraphSlam.
LidarGraphSlamUpdate
Diagnostics returned by LidarGraphSlam::update.
LoopClosure
An accepted loop closure edge.