rsasaki0109/lidar_slam_ros2

What it solves

Most LiDAR SLAM (Simultaneous Localization and Mapping) tools only provide a trajectory and a point cloud. This project bridges the gap between raw SLAM output and usable autonomous driving maps by generating Autoware-compatible map bundles, including point cloud maps, projector information, and automatically generated lanelet2 drivable lanes.

How it works

The system uses a two-stage pipeline: a frontend called RKO-LIO for LiDAR-inertial odometry and a backend called graph_based_slam for loop closure and graph optimization. It processes ROS 2 bags (containing LiDAR and IMU data) to create a globally refined map. It also includes specialized presets for challenging environments, such as radar fusion and gravity alignment to prevent "degeneracy collapse" in long, self-similar tunnels or foggy conditions. Additionally, it can project synchronized camera pixels onto the LiDAR geometry to create colorized 3D maps.

Who it’s for

This tool is designed for robotics engineers and autonomous vehicle developers who need to convert raw sensor data (rosbags) into high-definition maps that can be directly loaded into the Autoware ecosystem.

Highlights

• Autoware Integration: Directly outputs pointcloud_map/ and map_projector_info.yaml files that pass Autoware verification.
• Automated Lanelet Generation: Creates drivable lanelets from the SLAM trajectory for routing.
• Degeneracy Resilience: Specialized modes for tunnels and fog that prevent the mapping process from freezing or drifting.
• Deterministic Mapping: Ensures that offline runners produce identical trajectories and submaps across runs.
• Camera Fusion: Supports projecting RGB camera data onto point clouds for visually rich maps.
• Broad Sensor Support: Compatible with Ouster, Velodyne, RoboSense, and Livox (MID-360) sensors.