Try It in Your Browser¶
RKO-LIO, the LiDAR-inertial odometry behind lidarslam-map start, runs here as
WebAssembly. Press Start: the page downloads a 60-second clip of a public
Livox MID-360 drive (7.4 MB) and builds the map on your device, in a worker
thread. Nothing is uploaded.
Open the demo on its own page.
What runs¶
- The odometry is
rko_lio::driver::OfflineDriverwith the RKO-LIO core, the same C++ the ROS 2 node and thelidarslamPython package run, compiled with Emscripten. It is single-threaded: a serial stand-in for oneTBB (web/tbb_shim) keeps oneTBB's splitting, so deterministic sums are formed in the same order. - On the clip, the WebAssembly build gives the native build's trajectory: all
599 poses agree to the 9 printed decimals (
web/replay_node.jschecks this againstweb/demo_reference.tum, written by the native build). - The map is the accumulated deskewed scans at the odometry poses. The pose
graph, loop closure and the Autoware map bundle of the full pipeline are not
part of this page; use
pip install lidarslamor the ROS 2 packages for those.
The clip¶
docs/demo/clip_mid360_60s.lswc2.gz holds the first 60 s of
rosbag2_2024_04_16-14_17_01 from the public Livox MID-360 driving recording
(Zenodo 14841855, CC BY 4.0): the IMU
samples as recorded, and every fourth LiDAR point with coordinates rounded to
1 cm and point times to 2 µs. web/make_clip.py writes it from the bag. On
this thinned clip the odometry stays within 2 m of the raw trajectory of the
product demo (lidarslam-map demo, all points) over the same 60 s: 0.8 m on
average over 178 m.
Build the module yourself with web/build_wasm.sh; see web/README.md.