experiment on KITTI
illuosion opened this issue · 10 comments
If I want to do an image experiment on a KITTI dataset, where should I download the standard dataset which is applied to Swarm-SLAM?could u give a link?
and I have problem in graco， the graco_lidar.launch.py has been run 2h,is it possible?
Hi! Unfortunately, I don't have download links for those massive datasets. I rely on theirs authors for the dataset availability. So you will have to download, split and convert the KITTI dataset yourself : https://www.cvlibs.net/datasets/kitti/ , https://github.com/tomas789/kitti2bag , https://discourse.ros.org/t/rosbags-the-pure-python-library-for-everything-rosbag/20944
For the second problem, one or more nodes must have crashed. Please check if all ros2 nodes are still alive with ros2 node list, you can post the output here.
For the second problem, one or more nodes must have crashed. Please check if all ros2 nodes are still alive with
ros2 node list, you can post the output here.
btw ，how 2 visualize the odometry result and map while testing graco
the final output above
the final output above
should I change some params ?According to your screenshot one of the pose graph manager has crashed which explains the observed behavior. This is probably due to a out-of-memory error. Running multiple Swarm-SLAM agents + ROS 2 bags on a single computer requires a lot of RAM. I suggest that you reduce the rate at which the ROS 2 bags are played and the time offset between them https://github.com/lajoiepy/cslam_experiments/blob/9f72df6549478e55643e5c1ae566675d14faee13/launch/datasets_experiments/kitti_lidar.launch.py#L135 and https://github.com/lajoiepy/cslam_experiments/blob/9f72df6549478e55643e5c1ae566675d14faee13/launch/datasets_experiments/kitti_lidar.launch.py#L130
Alternatively, you can decrease the Swarm-SLAM performance by increasing the various loop periods available in the configuration files such as https://github.com/lajoiepy/cslam_experiments/blob/9f72df6549478e55643e5c1ae566675d14faee13/config/kitti_lidar.yaml#L15
The best solution is to run each agent on a different computer in the same network as this is the target design of Swarm-SLAM.
You can see the odometry results with ros2 topic echo /rX/odom where X is the robot ID.
Also you can visualize a live visualization of the pose graph estimates using https://github.com/lajoiepy/cslam_visualization
My server has 32 GB RAM，how the system performs if I use 2 robots
I want to clarify that the system does not shutdown on itself. It keeps running until you kill it. I strongly suggest that you run the cslam_visualization to see the results. You can also enable logging https://github.com/lajoiepy/cslam_experiments/blob/9f72df6549478e55643e5c1ae566675d14faee13/config/kitti_lidar.yaml#L34 and see the resulting files in your log_folder.