RAM filling up rapidly

[aaravrav142]

Hi I am testing FLOAM with the ouster 64 Lidar ROS dataset available from their website. I have a 32 GB RAM, running the ROS melodic. The FLOAM is quickly filling up all the RAM within minutes of running the node.

Can you suggest something to solve this?

Thanks

[Benjamin-Tan]

I am facing the same issue too, the huge memory consumption mainly comes from the odom estimation node.

[aaravrav142]

Hi did you find any solution for this problem?

[JanQuenzel]

I faced a similar issue while running floam on the Newer College dataset recorded with an Ouster OS1-64. The odometryEstimationNode has 2 queues ( pointCloudEdgeBuf and pointCloudSurfBuf ). The callbacks velodyneSurfHandler and velodyneEdgeHandler only push into these queues, while odom_estimation() retrieves the data. Once the odometry estimation can't process the data fast enough (>100ms) both queues fill up rapidly. For usage on a real robot this may lead to incorrect behaviour, since the current estimate gets more and more outdated. It is better to replace both queues with deques and push_back new messages and pop_front older messages if a certain maximum queue size is reached. Furthermore I would recommend to reduce all subscribers / publishers queue sizes down from 100 to maybe 5 or 10 too.

[masterhui]

Hey @JanQuenzel I am also facing issues with queues filling up with my Ouster OS0-64. Have you implemented the changes you suggested yet by any chance? Cheers

[masterhui]

OK I just saw commit 5f592a2106400b775e00ed2d04af177af973b1d9 in your forked floam repo. I'll check it out :-)