Skip to content

lidarslam_ros2 v0.2.1

This is the public v2 beta refresh release for lidarslam_ros2.

Date: 2026-03-28

Highlights

  • GNSS-aware graph optimization now uses covariance-based weighting and has real open-data validation for both direct NavSatFix bags and Applanix sidecar conversion
  • packet-path IMU deskew was hardened around PointCloud2.time handling and validated on real open data with a repeatable matrix report
  • save-time dynamic-object filtering now has cross-dataset validation on Leo Drive bag1 and bag6
  • classic-path fallback benchmarking now includes velocity-based odom-prior comparisons and a validation report that keeps dataset-specific knobs out of the public default
  • exploratory place-recognition work is now closed out in a single report: distance remains the public default, while Scan Context, BEV rerank, and SOLiD stay opt-in or experimental

Public Snapshot

  • fixed public entrypoint: bash scripts/run_autoware_quickstart.sh
  • map-authoring summary: python3 scripts/generate_map_authoring_report.py
  • place/classic closeout: python3 scripts/generate_exploration_closeout_report.py
  • NTU VIRAL current default APE RMSE: 0.952 m
  • NTU VIRAL best observed APE RMSE: 0.870 m
  • MID360 current default cross-validation APE RMSE: 3.641 m
  • MID360 best observed cross-validation APE RMSE: 3.590 m
  • comparison page: docs/comparison.md

Supported Scope

  • ROS 2 LiDAR SLAM with a non-GPL default path
  • pointcloud-map generation for Autoware-compatible workflows
  • map_projector_info.yaml output as Local without GNSS and LocalCartesian with a stable GNSS origin
  • release-readiness, map-authoring, dynamic-filter, and exploration-closeout reports from collected local artifacts

Known Limits

  • lanelet generation is out of scope
  • MID360 evidence is still cross-validation against GLIM, not ground truth
  • classic-path improvements remain fallback-path evidence, not the main public workflow
  • place-recognition descriptor experiments are documented, but the public default still relies on the distance-based path

Recommended Entry Points

bash scripts/run_autoware_quickstart.sh
bash scripts/run_rko_lio_graph_benchmark.sh
bash scripts/run_release_readiness_checks.sh --ape-threshold 0.10
python3 scripts/generate_map_authoring_report.py

Release Artifacts

For public distribution, publish at least:

  • docs/comparison.md
  • output/benchmark_summary.md
  • output/latest_report.html
  • output/v2_beta_readiness_20260324.md
  • output/stress_validation_report_20260325.md
  • output/map_authoring_report_20260326.md