IT Log

Record various IT issues and difficulties.

Using fastlio2 for state estimation, we have found that the global point cloud map topics it creates consume too much runtime memory, affecting the exploration area. How can this issue be addressed


To address the high memory consumption of the global point cloud map generated by FastLIO2, consider implementing the following strategies:

  1. Downsampling the Point Cloud: Reduce the resolution or density of the point cloud using techniques like voxel grid filtering or random sampling. This minimizes data size while retaining sufficient navigational accuracy.

  2. Optimizing Data Structures: Transition to more memory-efficient data structures such as k-d trees for spatial partitioning, which can reduce both computational and memory complexity.

  3. Adjusting FastLIO2 Parameters: Explore configuration options within FastLIO2, such as limiting map size or setting thresholds for data pruning, to manage memory usage effectively.

  4. Implementing Dynamic Maps: In dynamic environments, adopt a dynamic mapping approach where only relevant data is stored and outdated information is discarded, freeing up memory.

  5. Efficient Data Storage Formats: Investigate more efficient storage formats or compression methods for the point cloud, such as sparse representations or encoding techniques that reduce data size.

  6. Algorithmic Optimizations: Review FastLIO2’s source code to identify inefficient memory allocations or leaks. Use profiling tools to isolate resource-intensive components and optimize accordingly.

By systematically applying these strategies, you can mitigate high memory usage in FastLIO2, ensuring smoother operation and improved exploration capabilities for the robot.


, , , ,