🛢️ DSOL: Direct Sparse Odometry Lite

Build Directions for ERL

1. Recursively clone erl_car_localization repository in your catkin workspace.

2. Run Catkin build with Release version.

All the missing dependencies are built simultaneously with catkin cmake. These include abseil, fmt, glog, xtl, xtensor.

Run Instructions for ERL

Use dsol_ros to run realtime using data from ros topic, and dsol_data to run on datasets. Check the launch file to see the configurable parameters; topic names, camera intrinsics, frame names can be modified here.

Reference

Chao Qu, Shreyas S. Shivakumar, Ian D. Miller, Camillo J. Taylor

https://arxiv.org/abs/2203.08182

https://youtu.be/yunBYUACUdg

Datasets

VKITTI2 https://europe.naverlabs.com/research/computer-vision/proxy-virtual-worlds-vkitti-2/

KITTI Odom

TartanAir https://theairlab.org/tartanair-dataset/

Sample realsense data at

https://www.dropbox.com/s/bidng4gteeh8sx3/20220307_172336.bag?dl=0

https://www.dropbox.com/s/e8aefoji684dp3r/20220307_171655.bag?dl=0

Calib for realsense data is

393.4910888671875 393.4910888671875 318.6263122558594 240.12942504882812 0.095150406

Put this in calib.txt and put it in the same folder of the realsense dataset generated by the python file.

Build

This is a ros package, just put in a catkin workspace and build the workspace.

Run

Open rviz using the config in launch/dsol.rviz

roslaunch dsol dsol_data.launch

See launch files for more details on different datasets.

See config folder for details on configs.

To run multithread and show timing every 5 frames do

roslaunch dsol dsol_data.launch tbb:=1 log:=5

Dependencies

See CMakeLists.txt for dependencies.

Disclaimer

For reproducing the results in the paper, place use the iros22 branch.

This is the open-source version, advanced features are not included.