finds.dev← search

// the find

rsasaki0109/lidar_slam_ros2

★ 820 · HTML · BSD-2-Clause · updated Jun 2026

ROS 2 LiDAR SLAM for pointcloud-map authoring, benchmarking, and Autoware-compatible map workflows.

A ROS 2 LiDAR SLAM stack that goes further than most: it doesn't just produce a trajectory and point cloud, it produces a complete Autoware-ready map bundle with lanelet2, validates it, and dogs it end-to-end through AWSIM autonomous driving. Aimed at robotics engineers and AV developers who need production-grade maps, not research demos.

CI gates APE thresholds per dataset using real surveyed ground truth (total-station checkpoints, Leica prisms) — not just 'it compiles'. Deterministic offline replay is enforced with byte-identical trajectory checks across 3 runs, which is rare and genuinely useful for debugging regressions. The Docker one-liner that downloads a real bag and produces a loadable Autoware map is a good onboarding experience. GPL-free default workflow is a practical decision that matters for commercial use.

Accuracy numbers for larger outdoor loops (Stadtgarten, 1.6 m RMSE) are report-only and haven't graduated to hard CI gates — the README buries this distinction and optimistic readers will miss it. The lanelet2 auto-generation from SLAM trajectory is a convenience wrapper, not a real HD map; it produces drivable lanelets but not the lane semantics (speed limits, turn restrictions, traffic lights) that Autoware actually needs for real-world deployments. The RKO-LIO frontend is a git submodule, so the build chain and API surface are outside this project's control — a breaking upstream change could silently break everything here.

View on GitHub → Homepage ↗

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →